{ "count": 618, "note": "reference estimates only (source models, not measurements); excluded from record counts", "records": [ { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-phi-4-mini-m5-max-q4-k-m-ollama", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 Mini,3.8B,Q4_K_M,M5 Max,64,Ollama,168,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 168, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M4 Pro", "id": "llmcheck-phi-4-mini-m4-pro-q4-k-m-ollama", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 Mini,3.8B,Q4_K_M,M4 Pro,24,Ollama,86,None,2026-03", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 86, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M3", "id": "llmcheck-phi-4-mini-m3-q4-k-m-mlx", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 Mini,3.8B,Q4_K_M,M3,16,MLX,34,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 34, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M2", "id": "llmcheck-phi-4-mini-m2-q4-k-m-ollama", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 Mini,3.8B,Q4_K_M,M2,8,Ollama,34,None,2026-01", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 34, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M1", "id": "llmcheck-phi-4-mini-m1-q4-k-m-ollama", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 Mini,3.8B,Q4_K_M,M1,16,Ollama,24,None,2025-12", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 24, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-5-4b-m5-max-q4-k-m-mlx", "model": "Qwen 3.5 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 4B,4B,Q4_K_M,M5 Max,64,MLX,161,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4", "id": "llmcheck-qwen-3-5-4b-m4-q4-k-m-ollama", "model": "Qwen 3.5 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 4B,4B,Q4_K_M,M4,16,Ollama,39,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 39, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M2", "id": "llmcheck-qwen-3-4b-m2-q4-k-m-ollama", "model": "Qwen 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 4B,4B,Q4_K_M,M2,8,Ollama,33,None,2026-01", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 33, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-4b-m4-pro-q4-k-m-mlx", "model": "Qwen 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 4B,4B,Q4_K_M,M4 Pro,24,MLX,82,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 82, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-gemma-4-e2b-m5-max-q4-k-m-mlx", "model": "Gemma 4 E2B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "2.3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E2B,2.3B,Q4_K_M,M5 Max,128,MLX,261,None,2026-04", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 261, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Pro", "id": "llmcheck-gemma-4-e2b-m4-pro-q4-k-m-ollama", "model": "Gemma 4 E2B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "2.3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E2B,2.3B,Q4_K_M,M4 Pro,24,Ollama,147,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 147, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M3", "id": "llmcheck-gemma-4-e2b-m3-q4-k-m-ollama", "model": "Gemma 4 E2B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "2.3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E2B,2.3B,Q4_K_M,M3,16,Ollama,62,None,2026-04", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 62, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M1", "id": "llmcheck-gemma-4-e2b-m1-q4-k-m-ollama", "model": "Gemma 4 E2B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "2.3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E2B,2.3B,Q4_K_M,M1,8,Ollama,44,None,2026-04", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 44, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-gemma-4-e4b-m5-max-q4-k-m-mlx", "model": "Gemma 4 E4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E4B,4B,Q4_K_M,M5 Max,128,MLX,161,None,2026-04", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Pro", "id": "llmcheck-gemma-4-e4b-m5-pro-q4-k-m-ollama", "model": "Gemma 4 E4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E4B,4B,Q4_K_M,M5 Pro,24,Ollama,91,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Pro", "id": "llmcheck-gemma-4-e4b-m4-pro-q4-k-m-mlx", "model": "Gemma 4 E4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E4B,4B,Q4_K_M,M4 Pro,24,MLX,82,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 82, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M3", "id": "llmcheck-gemma-4-e4b-m3-q4-k-m-ollama", "model": "Gemma 4 E4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E4B,4B,Q4_K_M,M3,16,Ollama,33,None,2026-04", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 33, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M1", "id": "llmcheck-gemma-4-e4b-m1-q4-k-m-ollama", "model": "Gemma 4 E4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E4B,4B,Q4_K_M,M1,8,Ollama,23,None,2026-04", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-gemma-4-26b-a4b-m5-max-q4-k-m-mlx", "model": "Gemma 4 26B-A4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "26B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 26B-A4B,26B,Q4_K_M,M5 Max,128,MLX,106,None,2026-04", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 106, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Pro", "id": "llmcheck-gemma-4-26b-a4b-m5-pro-q4-k-m-ollama", "model": "Gemma 4 26B-A4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "26B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 26B-A4B,26B,Q4_K_M,M5 Pro,24,Ollama,72,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 72, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Max", "id": "llmcheck-gemma-4-26b-a4b-m4-max-q4-k-m-mlx", "model": "Gemma 4 26B-A4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "26B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 26B-A4B,26B,Q4_K_M,M4 Max,48,MLX,100,None,2026-04", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 100, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Pro", "id": "llmcheck-gemma-4-26b-a4b-m4-pro-q4-k-m-ollama", "model": "Gemma 4 26B-A4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "26B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 26B-A4B,26B,Q4_K_M,M4 Pro,24,Ollama,67,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 67, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-gemma-4-31b-m5-max-q4-k-m-mlx", "model": "Gemma-4-31B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "31B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 31B,31B,Q4_K_M,M5 Max,128,MLX,26,None,2026-04", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Max", "id": "llmcheck-gemma-4-31b-m4-max-q4-k-m-mlx", "model": "Gemma-4-31B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "31B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 31B,31B,Q4_K_M,M4 Max,48,MLX,23,None,2026-04", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Pro", "id": "llmcheck-gemma-4-31b-m4-pro-q4-k-m-ollama", "model": "Gemma-4-31B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "31B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 31B,31B,Q4_K_M,M4 Pro,24,Ollama,12,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-gemma-3-4b-m5-max-q4-k-m-ollama", "model": "Gemma 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 4B,4B,Q4_K_M,M5 Max,64,Ollama,161,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M1", "id": "llmcheck-gemma-3-4b-m1-q4-k-m-ollama", "model": "Gemma 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 4B,4B,Q4_K_M,M1,8,Ollama,23,None,2025-12", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3 Pro", "id": "llmcheck-gemma-3-4b-m3-pro-q4-k-m-mlx", "model": "Gemma 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 4B,4B,Q4_K_M,M3 Pro,18,MLX,48,None,2026-01", "ram_gb": 18, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 48, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-5-9b-m5-max-q4-k-m-ollama", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 9B,9B,Q4_K_M,M5 Max,64,Ollama,82,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 82, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4", "id": "llmcheck-qwen-3-5-9b-m4-q4-k-m-lm-studio", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 9B,9B,Q4_K_M,M4,16,LM Studio,18,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 18, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3", "id": "llmcheck-qwen-3-5-9b-m3-q4-k-m-ollama", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 9B,9B,Q4_K_M,M3,16,Ollama,15,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M1", "id": "llmcheck-qwen-3-5-9b-m1-q4-k-m-ollama", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 9B,9B,Q4_K_M,M1,16,Ollama,10,None,2025-12", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-8b-m5-max-q4-k-m-ollama", "model": "Qwen 3 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 8B,8B,Q4_K_M,M5 Max,128,Ollama,91,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-8b-m4-pro-q8-0-mlx", "model": "Qwen 3 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Qwen 3 8B,8B,Q8_0,M4 Pro,24,MLX,24,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 24, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M2", "id": "llmcheck-qwen-3-8b-m2-q4-k-m-ollama", "model": "Qwen 3 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 8B,8B,Q4_K_M,M2,16,Ollama,17,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-deepseek-r1-8b-m5-max-q4-k-m-ollama", "model": "DeepSeek R1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 8B,8B,Q4_K_M,M5 Max,64,Ollama,91,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4", "id": "llmcheck-deepseek-r1-8b-m4-q4-k-m-mlx", "model": "DeepSeek R1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 8B,8B,Q4_K_M,M4,16,MLX,20,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M2", "id": "llmcheck-deepseek-r1-8b-m2-q4-k-m-ollama", "model": "DeepSeek R1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 8B,8B,Q4_K_M,M2,16,Ollama,17,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-11", "hardware": "M1", "id": "llmcheck-deepseek-r1-8b-m1-q4-k-m-ollama", "model": "DeepSeek R1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 8B,8B,Q4_K_M,M1,16,Ollama,11,None,2025-11", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-mistral-7b-m5-max-q4-k-m-ollama", "model": "Mistral 7B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "7B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral 7B,7B,Q4_K_M,M5 Max,64,Ollama,102,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 102, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-mistral-7b-m4-pro-q4-k-m-mlx", "model": "Mistral 7B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "7B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral 7B,7B,Q4_K_M,M4 Pro,24,MLX,50,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 50, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3", "id": "llmcheck-mistral-7b-m3-q4-k-m-ollama", "model": "Mistral 7B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "7B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral 7B,7B,Q4_K_M,M3,16,Ollama,19,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 19, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M1", "id": "llmcheck-mistral-7b-m1-q4-k-m-ollama", "model": "Mistral 7B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "7B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral 7B,7B,Q4_K_M,M1,16,Ollama,13,None,2025-12", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 13, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-llama-3-1-8b-m5-max-q4-k-m-mlx", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 8B,8B,Q4_K_M,M5 Max,128,MLX,91,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4", "id": "llmcheck-llama-3-1-8b-m4-q4-k-m-ollama", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 8B,8B,Q4_K_M,M4,16,Ollama,20,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3 Pro", "id": "llmcheck-llama-3-1-8b-m3-pro-q4-k-m-ollama", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 8B,8B,Q4_K_M,M3 Pro,18,Ollama,25,None,2026-01", "ram_gb": 18, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M1", "id": "llmcheck-llama-3-1-8b-m1-q4-k-m-ollama", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 8B,8B,Q4_K_M,M1,16,Ollama,11,None,2025-12", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-ministral-8b-m5-max-q4-k-m-ollama", "model": "Ministral 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ministral 8B,8B,Q4_K_M,M5 Max,64,Ollama,91,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3", "id": "llmcheck-ministral-8b-m3-q4-k-m-lm-studio", "model": "Ministral 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ministral 8B,8B,Q4_K_M,M3,16,LM Studio,17,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-gemma-3-12b-m5-max-q4-k-m-ollama", "model": "Gemma 3 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 12B,12B,Q4_K_M,M5 Max,64,Ollama,64,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 64, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-gemma-3-12b-m4-pro-q4-k-m-mlx", "model": "Gemma 3 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 12B,12B,Q4_K_M,M4 Pro,24,MLX,30,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3", "id": "llmcheck-gemma-3-12b-m3-q4-k-m-ollama", "model": "Gemma 3 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 12B,12B,Q4_K_M,M3,16,Ollama,11,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M2", "id": "llmcheck-gemma-3-12b-m2-q4-k-m-ollama", "model": "Gemma 3 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 12B,12B,Q4_K_M,M2,16,Ollama,11,None,2025-12", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-14b-m5-max-q4-k-m-ollama", "model": "Qwen 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 14B,14B,Q4_K_M,M5 Max,64,Ollama,55,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 55, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-14b-m4-pro-q4-k-m-lm-studio", "model": "Qwen 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 14B,14B,Q4_K_M,M4 Pro,24,LM Studio,26,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3", "id": "llmcheck-qwen-3-14b-m3-q4-k-m-ollama", "model": "Qwen 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 14B,14B,Q4_K_M,M3,16,Ollama,10,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-phi-4-14b-m5-max-q4-k-m-mlx", "model": "Phi-4 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 14B,14B,Q4_K_M,M5 Max,64,MLX,55,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 55, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4", "id": "llmcheck-phi-4-14b-m4-q4-k-m-ollama", "model": "Phi-4 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 14B,14B,Q4_K_M,M4,16,Ollama,12,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M2", "id": "llmcheck-phi-4-14b-m2-q4-k-m-mlx", "model": "Phi-4 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 14B,14B,Q4_K_M,M2,16,MLX,10,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-ministral-3-14b-m5-max-q4-k-m-ollama", "model": "Ministral 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ministral 3 14B,14B,Q4_K_M,M5 Max,64,Ollama,55,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 55, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-ministral-3-14b-m4-pro-q4-k-m-ollama", "model": "Ministral 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ministral 3 14B,14B,Q4_K_M,M4 Pro,24,Ollama,26,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-gemma-3-27b-m5-max-q4-k-m-ollama", "model": "Gemma 3 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 27B,27B,Q4_K_M,M5 Max,64,Ollama,30,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-gemma-3-27b-m4-pro-q4-k-m-lm-studio", "model": "Gemma 3 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 27B,27B,Q4_K_M,M4 Pro,24,LM Studio,14,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 14, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Max", "id": "llmcheck-gemma-3-27b-m4-max-q4-k-m-mlx", "model": "Gemma 3 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 27B,27B,Q4_K_M,M4 Max,48,MLX,27,None,2026-02", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 27, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-30b-a3b-m5-max-q4-k-m-ollama", "model": "Qwen 3 30B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 30B-A3B,30B,Q4_K_M,M5 Max,64,Ollama,116,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 116, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-30b-a3b-m4-pro-q4-k-m-mlx", "model": "Qwen 3 30B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 30B-A3B,30B,Q4_K_M,M4 Pro,24,MLX,76,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 76, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3 Max", "id": "llmcheck-qwen-3-30b-a3b-m3-max-q4-k-m-ollama", "model": "Qwen 3 30B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 30B-A3B,30B,Q4_K_M,M3 Max,36,Ollama,94,None,2026-01", "ram_gb": 36, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 94, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-5-35b-a3b-m5-max-q4-k-m-mlx", "model": "Qwen 3.5 35B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 35B-A3B,35B,Q4_K_M,M5 Max,64,MLX,117,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 117, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Max", "id": "llmcheck-qwen-3-5-35b-a3b-m4-max-q4-k-m-ollama", "model": "Qwen 3.5 35B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 35B-A3B,35B,Q4_K_M,M4 Max,48,Ollama,111,None,2026-02", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 111, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-32b-m5-max-q4-k-m-ollama", "model": "Qwen 3 32B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "32B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 32B,32B,Q4_K_M,M5 Max,128,Ollama,25,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Max", "id": "llmcheck-qwen-3-32b-m4-max-q4-k-m-mlx", "model": "Qwen 3 32B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "32B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 32B,32B,Q4_K_M,M4 Max,64,MLX,23,None,2026-02", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-32b-m4-pro-q4-k-m-ollama", "model": "Qwen 3 32B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "32B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 32B,32B,Q4_K_M,M4 Pro,32,Ollama,12,None,2026-01", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-deepseek-r1-32b-m5-max-q4-k-m-ollama", "model": "DeepSeek R1 32B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "32B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 32B,32B,Q4_K_M,M5 Max,64,Ollama,25,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M4 Max", "id": "llmcheck-deepseek-r1-32b-m4-max-q4-k-m-lm-studio", "model": "DeepSeek R1 32B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "32B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 32B,32B,Q4_K_M,M4 Max,48,LM Studio,23,None,2026-01", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M3 Max", "id": "llmcheck-deepseek-r1-32b-m3-max-q4-k-m-ollama", "model": "DeepSeek R1 32B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "32B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 32B,32B,Q4_K_M,M3 Max,36,Ollama,17,None,2025-12", "ram_gb": 36, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-llama-3-3-70b-m5-max-q4-k-m-mlx", "model": "Llama 3.3 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.3 70B,70B,Q4_K_M,M5 Max,128,MLX,12,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-deepseek-r1-70b-m5-max-q4-k-m-ollama", "model": "DeepSeek R1 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 70B,70B,Q4_K_M,M5 Max,128,Ollama,12,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-2-5-72b-m5-max-q4-k-m-ollama", "model": "Qwen 2.5 72B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "72B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 2.5 72B,72B,Q4_K_M,M5 Max,128,Ollama,12,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-llama-4-scout-m5-max-q4-k-m-mlx", "model": "Llama 4 Scout", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "109B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 4 Scout,109B,Q4_K_M,M5 Max,128,MLX,19,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 19, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-mistral-7b-m5-max-q8-0-mlx", "model": "Mistral 7B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "7B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Mistral 7B,7B,Q8_0,M5 Max,64,MLX,59,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 59, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-mistral-7b-m4-pro-q8-0-ollama", "model": "Mistral 7B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "7B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Mistral 7B,7B,Q8_0,M4 Pro,24,Ollama,28,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 28, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-5-9b-m5-max-q8-0-mlx", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Qwen 3.5 9B,9B,Q8_0,M5 Max,128,MLX,47,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 47, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-llama-3-1-8b-m5-max-q8-0-ollama", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Llama 3.1 8B,8B,Q8_0,M5 Max,64,Ollama,52,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 52, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-14b-m5-max-q8-0-mlx", "model": "Qwen 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Qwen 3 14B,14B,Q8_0,M5 Max,128,MLX,31,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 31, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-phi-4-mini-m5-max-q8-0-mlx", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Phi-4 Mini,3.8B,Q8_0,M5 Max,64,MLX,102,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 102, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Max", "id": "llmcheck-phi-4-mini-m4-max-q4-k-m-mlx", "model": "Phi-4 Mini", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 Mini,3.8B,Q4_K_M,M4 Max,48,MLX,153,None,2026-02", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 153, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-gemma-3-4b-m4-pro-q8-0-mlx", "model": "Gemma 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "Gemma 3 4B,4B,Q8_0,M4 Pro,24,MLX,47,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 47, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3 Max", "id": "llmcheck-qwen-3-5-35b-a3b-m3-max-q4-k-m-ollama", "model": "Qwen 3.5 35B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 35B-A3B,35B,Q4_K_M,M3 Max,96,Ollama,96,None,2026-01", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 96, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-deepseek-r1-8b-m5-max-q8-0-mlx", "model": "DeepSeek R1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "DeepSeek R1 8B,8B,Q8_0,M5 Max,64,MLX,52,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 52, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-8b-m4-pro-q4-k-m-ollama", "model": "Qwen 3 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 8B,8B,Q4_K_M,M4 Pro,24,Ollama,44,None,2026-02", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 44, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M1", "id": "llmcheck-qwen-3-8b-m1-q4-k-m-ollama", "model": "Qwen 3 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 8B,8B,Q4_K_M,M1,16,Ollama,11,None,2025-12", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-02", "hardware": "M4", "id": "llmcheck-ministral-8b-m4-q4-k-m-mlx", "model": "Ministral 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ministral 8B,8B,Q4_K_M,M4,16,MLX,20,None,2026-02", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3", "id": "llmcheck-ministral-3-14b-m3-q4-k-m-lm-studio", "model": "Ministral 3 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ministral 3 14B,14B,Q4_K_M,M3,16,LM Studio,10,None,2026-01", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-01", "hardware": "M3 Max", "id": "llmcheck-gemma-3-27b-m3-max-q4-k-m-mlx", "model": "Gemma 3 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 27B,27B,Q4_K_M,M3 Max,96,MLX,20,None,2026-01", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2025-12", "hardware": "M2", "id": "llmcheck-llama-3-1-8b-m2-q4-k-m-ollama", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 8B,8B,Q4_K_M,M2,8,Ollama,17,None,2025-12", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-5-9b-m4-pro-q4-k-m-mlx", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 9B,9B,Q4_K_M,M4 Pro,24,MLX,39,None,2026-03", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 39, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-qwen-3-4b-m5-max-q4-k-m-ollama", "model": "Qwen 3 4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3 4B,4B,Q4_K_M,M5 Max,64,Ollama,161,None,2026-03", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Max", "id": "llmcheck-qwen-3-6-35b-a3b-m4-max-q4-k-m-mlx", "model": "Qwen 3.6-35B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-35B-A3B,35B,Q4_K_M,M4 Max,48,MLX,120,None,2026-04", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 120, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-6-35b-a3b-m4-pro-q4-k-m-ollama", "model": "Qwen 3.6-35B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-35B-A3B,35B,Q4_K_M,M4 Pro,24,Ollama,86,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 86, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-mistral-small-4-m5-max-q4-k-m-mlx", "model": "Mistral Small 4", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "119B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral Small 4,119B,Q4_K_M,M5 Max,128,MLX,46,None,2026-04", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 46, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-qwen3-235b-a22b-m5-max-q4-k-m-mlx", "model": "Qwen3-235B-A22B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "235B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3-235B-A22B,235B,Q4_K_M,M5 Max,128,MLX,15,None,2026-04", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M5 Max", "id": "llmcheck-nemotron-cascade-2-m5-max-q4-k-m-ollama", "model": "Nemotron-Cascade 2", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nemotron-Cascade 2,30B,Q4_K_M,M5 Max,64,Ollama,117,None,2026-04", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 117, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Max", "id": "llmcheck-nemotron-cascade-2-m4-max-q4-k-m-mlx", "model": "Nemotron-Cascade 2", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nemotron-Cascade 2,30B,Q4_K_M,M4 Max,48,MLX,111,None,2026-04", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 111, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-04", "hardware": "M4 Pro", "id": "llmcheck-nemotron-cascade-2-m4-pro-q4-k-m-ollama", "model": "Nemotron-Cascade 2", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nemotron-Cascade 2,30B,Q4_K_M,M4 Pro,24,Ollama,77,None,2026-04", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 77, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M5 Max", "id": "llmcheck-mistral-small-3-2-24b-m5-max-q4-k-m-mlx", "model": "Mistral Small 3.2 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral Small 3.2 24B,24B,Q4_K_M,M5 Max,128,MLX,33,None,2026-05", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 33, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M4 Max", "id": "llmcheck-mistral-small-3-2-24b-m4-max-q4-k-m-mlx", "model": "Mistral Small 3.2 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral Small 3.2 24B,24B,Q4_K_M,M4 Max,48,MLX,30,None,2026-05", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M4 Pro", "id": "llmcheck-mistral-small-3-2-24b-m4-pro-q4-k-m-ollama", "model": "Mistral Small 3.2 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral Small 3.2 24B,24B,Q4_K_M,M4 Pro,32,Ollama,15,None,2026-05", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M5 Max", "id": "llmcheck-hermes-4-70b-m5-max-q4-k-m-mlx", "model": "Hermes 4 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Hermes 4 70B,70B,Q4_K_M,M5 Max,128,MLX,12,None,2026-05", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M4 Max", "id": "llmcheck-hermes-4-70b-m4-max-q4-k-m-mlx", "model": "Hermes 4 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Hermes 4 70B,70B,Q4_K_M,M4 Max,128,MLX,11,None,2026-05", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M5 Max", "id": "llmcheck-smollm3-3b-m5-max-q4-k-m-mlx", "model": "SmolLM3 3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "SmolLM3 3B,3B,Q4_K_M,M5 Max,64,MLX,199,None,2026-05", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 199, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M4 Pro", "id": "llmcheck-smollm3-3b-m4-pro-q4-k-m-ollama", "model": "SmolLM3 3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "SmolLM3 3B,3B,Q4_K_M,M4 Pro,24,Ollama,106,None,2026-05", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 106, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M3", "id": "llmcheck-smollm3-3b-m3-q4-k-m-ollama", "model": "SmolLM3 3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "SmolLM3 3B,3B,Q4_K_M,M3,16,Ollama,43,None,2026-05", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 43, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M2", "id": "llmcheck-smollm3-3b-m2-q4-k-m-ollama", "model": "SmolLM3 3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "SmolLM3 3B,3B,Q4_K_M,M2,8,Ollama,43,None,2026-05", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 43, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M1", "id": "llmcheck-smollm3-3b-m1-q4-k-m-ollama", "model": "SmolLM3 3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "SmolLM3 3B,3B,Q4_K_M,M1,8,Ollama,30,None,2026-05", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M5 Max", "id": "llmcheck-devstral-small-24b-m5-max-q4-k-m-mlx", "model": "Devstral Small 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Devstral Small 24B,24B,Q4_K_M,M5 Max,128,MLX,33,None,2026-05", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 33, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M4 Max", "id": "llmcheck-devstral-small-24b-m4-max-q4-k-m-mlx", "model": "Devstral Small 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Devstral Small 24B,24B,Q4_K_M,M4 Max,48,MLX,30,None,2026-05", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-05", "hardware": "M4 Pro", "id": "llmcheck-devstral-small-24b-m4-pro-q4-k-m-ollama", "model": "Devstral Small 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Devstral Small 24B,24B,Q4_K_M,M4 Pro,32,Ollama,15,None,2026-05", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-07", "hardware": "M5 Max", "id": "llmcheck-glm-4-5-air-m5-max-q4-k-m-mlx", "model": "GLM-4.5-Air", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "106B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-4.5-Air,106B,Q4_K_M,M5 Max,128,MLX,27,None,2026-07", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 27, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-07", "hardware": "M4 Max", "id": "llmcheck-glm-4-5-air-m4-max-q4-k-m-mlx", "model": "GLM-4.5-Air", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "106B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-4.5-Air,106B,Q4_K_M,M4 Max,128,MLX,24,None,2026-07", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 24, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-qwen-3-6-27b-m5-max-q4-k-m-mlx", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M5 Max,64,MLX,30,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Max", "id": "llmcheck-qwen-3-6-27b-m4-max-q4-k-m-mlx", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M4 Max,64,MLX,27,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 27, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M3 Max", "id": "llmcheck-qwen-3-6-27b-m3-max-q4-k-m-ollama", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M3 Max,64,Ollama,20,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Pro", "id": "llmcheck-qwen-3-6-27b-m5-pro-q4-k-m-mlx", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M5 Pro,24,MLX,15,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Pro", "id": "llmcheck-qwen-3-6-27b-m4-pro-q4-k-m-ollama", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M4 Pro,24,Ollama,14,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 14, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-kat-coder-v2-5-m5-max-q4-k-m-mlx", "model": "KAT-Coder-V2.5", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "KAT-Coder-V2.5,35B-A3B,Q4_K_M,M5 Max,64,MLX,117,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 117, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Max", "id": "llmcheck-kat-coder-v2-5-m4-max-q4-k-m-mlx", "model": "KAT-Coder-V2.5", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "KAT-Coder-V2.5,35B-A3B,Q4_K_M,M4 Max,48,MLX,111,None,2026-08", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 111, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Pro", "id": "llmcheck-kat-coder-v2-5-m4-pro-q4-k-m-ollama", "model": "KAT-Coder-V2.5", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "KAT-Coder-V2.5,35B-A3B,Q4_K_M,M4 Pro,24,Ollama,77,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 77, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-nemotron-3-5-lightning-m5-max-q4-k-m-mlx", "model": "Nemotron 3.5 Lightning", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nemotron 3.5 Lightning,30B-A3B,Q4_K_M,M5 Max,64,MLX,103,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 103, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Max", "id": "llmcheck-nemotron-3-5-lightning-m4-max-q4-k-m-mlx", "model": "Nemotron 3.5 Lightning", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nemotron 3.5 Lightning,30B-A3B,Q4_K_M,M4 Max,48,MLX,97,None,2026-08", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 97, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Pro", "id": "llmcheck-nemotron-3-5-lightning-m4-pro-q4-k-m-ollama", "model": "Nemotron 3.5 Lightning", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "30B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nemotron 3.5 Lightning,30B-A3B,Q4_K_M,M4 Pro,24,Ollama,64,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 64, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-laguna-s-2-1-m5-max-q4-k-m-mlx", "model": "Laguna S 2.1", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "118B-A8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Laguna S 2.1,118B-A8B,Q4_K_M,M5 Max,128,MLX,38,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 38, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Max", "id": "llmcheck-laguna-s-2-1-m4-max-q4-k-m-mlx", "model": "Laguna S 2.1", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "118B-A8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Laguna S 2.1,118B-A8B,Q4_K_M,M4 Max,128,MLX,35,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 35, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-laguna-xs-2-1-m5-max-q4-k-m-mlx", "model": "Laguna XS 2.1", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "33B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Laguna XS 2.1,33B-A3B,Q4_K_M,M5 Max,64,MLX,117,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 117, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Pro", "id": "llmcheck-laguna-xs-2-1-m4-pro-q4-k-m-ollama", "model": "Laguna XS 2.1", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "33B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Laguna XS 2.1,33B-A3B,Q4_K_M,M4 Pro,24,Ollama,77,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 77, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-inkling-small-m5-max-q2-k-mlx", "model": "Inkling-Small", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "276B-A12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q2_K", "quote": "Inkling-Small,276B-A12B,Q2_K,M5 Max,128,MLX,42,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 42, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-bonsai-27b-m5-max-1-bit-mlx", "model": "Bonsai 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "1-bit", "quote": "Bonsai 27B,27B,1-bit,M5 Max,64,MLX,30,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4", "id": "llmcheck-bonsai-27b-m4-1-bit-mlx", "model": "Bonsai 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "1-bit", "quote": "Bonsai 27B,27B,1-bit,M4,16,MLX,6,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 6, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2", "id": "llmcheck-bonsai-27b-m2-1-bit-mlx", "model": "Bonsai 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "1-bit", "quote": "Bonsai 27B,27B,1-bit,M2,8,MLX,5,None,2026-08", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 5, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-nanbeige4-2-3b-m5-max-q4-k-m-mlx", "model": "Nanbeige4.2-3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nanbeige4.2-3B,3B,Q4_K_M,M5 Max,64,MLX,199,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 199, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4", "id": "llmcheck-nanbeige4-2-3b-m4-q4-k-m-ollama", "model": "Nanbeige4.2-3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nanbeige4.2-3B,3B,Q4_K_M,M4,16,Ollama,51,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 51, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2", "id": "llmcheck-nanbeige4-2-3b-m2-q4-k-m-ollama", "model": "Nanbeige4.2-3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Nanbeige4.2-3B,3B,Q4_K_M,M2,8,Ollama,43,None,2026-08", "ram_gb": 8, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 43, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-apertus-1-5-8b-m5-max-q4-k-m-mlx", "model": "Apertus 1.5 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Apertus 1.5 8B,8B,Q4_K_M,M5 Max,64,MLX,91,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4", "id": "llmcheck-apertus-1-5-8b-m4-q4-k-m-ollama", "model": "Apertus 1.5 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Apertus 1.5 8B,8B,Q4_K_M,M4,16,Ollama,20,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-apertus-1-5-70b-m5-max-q4-k-m-mlx", "model": "Apertus 1.5 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Apertus 1.5 70B,70B,Q4_K_M,M5 Max,128,MLX,12,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-glm-4-7-flash-m5-max-q4-k-m-mlx", "model": "GLM-4.7-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "31B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-4.7-Flash,31B,Q4_K_M,M5 Max,64,MLX,26,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Max", "id": "llmcheck-glm-4-7-flash-m4-max-q4-k-m-mlx", "model": "GLM-4.7-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "31B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-4.7-Flash,31B,Q4_K_M,M4 Max,48,MLX,23,None,2026-08", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Pro", "id": "llmcheck-glm-4-7-flash-m4-pro-q4-k-m-ollama", "model": "GLM-4.7-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "31B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-4.7-Flash,31B,Q4_K_M,M4 Pro,24,Ollama,12,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M1 Ultra", "id": "llmcheck-llama-3-3-70b-m1-ultra-q4-k-m-ollama", "model": "Llama 3.3 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.3 70B,70B,Q4_K_M,M1 Ultra,128,Ollama,15,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M1 Ultra", "id": "llmcheck-deepseek-r1-70b-m1-ultra-q4-k-m-ollama", "model": "DeepSeek R1 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 70B,70B,Q4_K_M,M1 Ultra,128,Ollama,15,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M1 Ultra", "id": "llmcheck-qwen-3-6-27b-m1-ultra-q4-k-m-mlx", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M1 Ultra,64,MLX,38,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 38, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M1 Ultra", "id": "llmcheck-gemma-4-26b-a4b-m1-ultra-q4-k-m-ollama", "model": "Gemma 4 26B-A4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "26B-A4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 26B-A4B,26B-A4B,Q4_K_M,M1 Ultra,64,Ollama,119,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 119, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2 Max", "id": "llmcheck-llama-3-3-70b-m2-max-q4-k-m-ollama", "model": "Llama 3.3 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.3 70B,70B,Q4_K_M,M2 Max,96,Ollama,8,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 8, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2 Max", "id": "llmcheck-qwen-3-6-35b-a3b-m2-max-q4-k-m-mlx", "model": "Qwen 3.6-35B-A3B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-35B-A3B,35B-A3B,Q4_K_M,M2 Max,64,MLX,105,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 105, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2 Max", "id": "llmcheck-qwen-3-6-27b-m2-max-q4-k-m-mlx", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M2 Max,32,MLX,20,None,2026-08", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2 Max", "id": "llmcheck-gemma-3-27b-m2-max-q4-k-m-ollama", "model": "Gemma 3 27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 3 27B,27B,Q4_K_M,M2 Max,32,Ollama,20,None,2026-08", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2 Max", "id": "llmcheck-llama-3-1-8b-m2-max-q4-k-m-ollama", "model": "Llama 3.1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 8B,8B,Q4_K_M,M2 Max,32,Ollama,62,None,2026-08", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 62, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M2 Max", "id": "llmcheck-mistral-small-3-2-24b-m2-max-q4-k-m-ollama", "model": "Mistral Small 3.2 24B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "24B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral Small 3.2 24B,24B,Q4_K_M,M2 Max,32,Ollama,22,None,2026-08", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 22, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M3 Ultra", "id": "llmcheck-llama-3-3-70b-m3-ultra-q4-k-m-ollama", "model": "Llama 3.3 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.3 70B,70B,Q4_K_M,M3 Ultra,256,Ollama,16,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 16, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M3 Ultra", "id": "llmcheck-deepseek-v4-flash-m3-ultra-q4-k-m-mlx", "model": "DeepSeek-V4-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "284B-A13B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek V4 Flash,284B-A13B,Q4_K_M,M3 Ultra,256,MLX,32,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 32, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M3 Ultra", "id": "llmcheck-inkling-small-m3-ultra-q4-k-m-mlx", "model": "Inkling-Small", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "276B-A12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Inkling-Small,276B-A12B,Q4_K_M,M3 Ultra,512,MLX,35,None,2026-08", "ram_gb": 512, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 35, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M3 Ultra", "id": "llmcheck-qwen3-235b-a22b-m3-ultra-q4-k-m-mlx", "model": "Qwen3-235B-A22B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "235B-A22B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3-235B-A22B,235B-A22B,Q4_K_M,M3 Ultra,256,MLX,20,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M3 Ultra", "id": "llmcheck-hunyuan-hy3-m3-ultra-q4-k-m-lm-studio", "model": "Hunyuan Hy3", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "295B-A21B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Hunyuan Hy3,295B-A21B,Q4_K_M,M3 Ultra,256,LM Studio,21,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 21, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Max", "id": "llmcheck-qwen3-coder-next-m5-max-q4-k-m-mlx", "model": "Qwen3-Coder-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "80B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3-Coder-Next,80B-A3B,Q4_K_M,M5 Max,64,MLX,84,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 84, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M4 Max", "id": "llmcheck-qwen3-coder-next-m4-max-q4-k-m-mlx", "model": "Qwen3-Coder-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "80B-A3B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3-Coder-Next,80B-A3B,Q4_K_M,M4 Max,64,MLX,77,None,2026-08", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 77, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-qwen3-8-27b-m6-q4-k-m-mlx", "model": "Qwen3.8-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-27B,27.8B,Q4_K_M,M6,24,MLX,8,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 8, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-qwen-3-6-27b-m6-q4-k-m-mlx", "model": "Qwen 3.6-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.6-27B,27B,Q4_K_M,M6,24,MLX,9,None,2026-08", "ram_gb": 24, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 9, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-kat-coder-v2-5-m6-q4-k-m-mlx", "model": "KAT-Coder-V2.5", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "35B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "KAT-Coder-V2.5,35B,Q4_K_M,M6,32,MLX,56,None,2026-08", "ram_gb": 32, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 56, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-deepseek-r1-8b-m6-q4-k-m-mlx", "model": "DeepSeek R1 8B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 8B,8B,Q4_K_M,M6,16,MLX,28,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 28, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-qwen-3-5-9b-m6-q4-k-m-mlx", "model": "Qwen 3.5 9B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "9B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 9B,9B,Q4_K_M,M6,16,MLX,25,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-phi-4-14b-m6-q4-k-m-mlx", "model": "Phi-4 14B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "14B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Phi-4 14B,14B,Q4_K_M,M6,16,MLX,16,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 16, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-gemma-4-e4b-m6-q4-k-m-mlx", "model": "Gemma 4 E4B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "4B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 E4B,4B,Q4_K_M,M6,16,MLX,54,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 54, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M6", "id": "llmcheck-lfm2-5-2-6b-m6-q4-k-m-mlx", "model": "LFM2.5-2.6B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "2.6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "LFM2.5-2.6B,2.6B,Q4_K_M,M6,16,MLX,79,None,2026-08", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 79, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-deepseek-v4-flash-m5-ultra-q4-k-m-mlx", "model": "DeepSeek-V4-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "284B-A13B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek V4 Flash,284B-A13B,Q4_K_M,M5 Ultra,256,MLX,45,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 45, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-inkling-small-m5-ultra-q4-k-m-mlx", "model": "Inkling-Small", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "276B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Inkling-Small,276B,Q4_K_M,M5 Ultra,256,MLX,49,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 49, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-qwen3-235b-a22b-m5-ultra-q4-k-m-mlx", "model": "Qwen3-235B-A22B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "235B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3-235B-A22B,235B,Q4_K_M,M5 Ultra,256,MLX,29,None,2026-08", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 29, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-mistral-small-4-m5-ultra-q4-k-m-mlx", "model": "Mistral Small 4", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "119B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Mistral Small 4,119B,Q4_K_M,M5 Ultra,96,MLX,79,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 79, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-llama-3-3-70b-m5-ultra-q4-k-m-mlx", "model": "Llama 3.3 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.3 70B,70B,Q4_K_M,M5 Ultra,96,MLX,23,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-deepseek-r1-70b-m5-ultra-q4-k-m-mlx", "model": "DeepSeek R1 70B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "70B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek R1 70B,70B,Q4_K_M,M5 Ultra,96,MLX,23,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-qwen3-coder-next-m5-ultra-q4-k-m-mlx", "model": "Qwen3-Coder-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "80B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3-Coder-Next,80B,Q4_K_M,M5 Ultra,96,MLX,130,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 130, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-glm-4-5-air-m5-ultra-q4-k-m-mlx", "model": "GLM-4.5-Air", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "106B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-4.5-Air,106B,Q4_K_M,M5 Ultra,96,MLX,49,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 49, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-laguna-s-2-1-m5-ultra-q4-k-m-mlx", "model": "Laguna S 2.1", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "118B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Laguna S 2.1,118B,Q4_K_M,M5 Ultra,96,MLX,68,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 68, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-llama-3-1-405b-m5-ultra-q4-k-m-mlx", "model": "Llama 3.1 405B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "405B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Llama 3.1 405B,405B,Q4_K_M,M5 Ultra,512,MLX,4,None,2026-08", "ram_gb": 512, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 4, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-qwen-2-5-72b-m5-ultra-q4-k-m-mlx", "model": "Qwen 2.5 72B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "72B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 2.5 72B,72B,Q4_K_M,M5 Ultra,96,MLX,23,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Max", "id": "llmcheck-glm-5-3-flash-m5-max-iq3-xxs-mlx", "model": "GLM-5.3-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "320B-A18B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "IQ3_XXS", "quote": "GLM-5.3-Flash,320B-A18B,IQ3_XXS,M5 Max,128,MLX,27,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 27, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Ultra", "id": "llmcheck-glm-5-3-flash-m3-ultra-q4-k-m-mlx", "model": "GLM-5.3-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "320B-A18B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-5.3-Flash,320B-A18B,Q4_K_M,M3 Ultra,256,MLX,24,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 24, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Ultra", "id": "llmcheck-glm-5-3-flash-m5-ultra-q4-k-m-mlx", "model": "GLM-5.3-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "320B-A18B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-5.3-Flash,320B-A18B,Q4_K_M,M5 Ultra,256,MLX,35,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 35, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Max", "id": "llmcheck-qwen3-8-flash-next-m5-max-q4-k-m-mlx", "model": "Qwen3.8-Flash-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "125B-A6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-Flash-Next,125B-A6B,Q4_K_M,M5 Max,128,MLX,49,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 49, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M4 Max", "id": "llmcheck-qwen3-8-flash-next-m4-max-q4-k-m-mlx", "model": "Qwen3.8-Flash-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "125B-A6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-Flash-Next,125B-A6B,Q4_K_M,M4 Max,128,MLX,44,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 44, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Ultra", "id": "llmcheck-qwen3-8-flash-next-m3-ultra-q4-k-m-mlx", "model": "Qwen3.8-Flash-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "125B-A6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-Flash-Next,125B-A6B,Q4_K_M,M3 Ultra,256,MLX,62,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 62, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Ultra", "id": "llmcheck-qwen3-8-flash-next-m5-ultra-q4-k-m-mlx", "model": "Qwen3.8-Flash-Next", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "125B-A6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-Flash-Next,125B-A6B,Q4_K_M,M5 Ultra,256,MLX,84,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 84, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Max", "id": "llmcheck-deepseek-v4-flash-vision-exp-m5-max-q2-k-llama-cpp", "model": "DeepSeek V4 Flash Vision-Exp", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "305B-A13B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q2_K", "quote": "DeepSeek V4 Flash Vision-Exp,305B-A13B,Q2_K,M5 Max,128,llama.cpp,39,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 39, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Ultra", "id": "llmcheck-deepseek-v4-flash-vision-exp-m3-ultra-q4-k-m-llama-cpp", "model": "DeepSeek V4 Flash Vision-Exp", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "305B-A13B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek V4 Flash Vision-Exp,305B-A13B,Q4_K_M,M3 Ultra,256,llama.cpp,32,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 32, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Ultra", "id": "llmcheck-deepseek-v4-flash-vision-exp-m5-ultra-q4-k-m-llama-cpp", "model": "DeepSeek V4 Flash Vision-Exp", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "305B-A13B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DeepSeek V4 Flash Vision-Exp,305B-A13B,Q4_K_M,M5 Ultra,256,llama.cpp,46,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 46, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Max", "id": "llmcheck-qwen3-8-27b-m5-max-q4-k-m-mlx", "model": "Qwen3.8-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-27B,27.8B,Q4_K_M,M5 Max,128,MLX,29,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 29, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M4 Max", "id": "llmcheck-qwen3-8-27b-m4-max-q4-k-m-mlx", "model": "Qwen3.8-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-27B,27.8B,Q4_K_M,M4 Max,64,MLX,26,None,2026-09", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Max", "id": "llmcheck-qwen3-8-27b-m3-max-q4-k-m-mlx", "model": "Qwen3.8-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-27B,27.8B,Q4_K_M,M3 Max,64,MLX,19,None,2026-09", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 19, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M4 Pro", "id": "llmcheck-qwen3-8-27b-m4-pro-q4-k-m-ollama", "model": "Qwen3.8-27B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "27.8B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen3.8-27B,27.8B,Q4_K_M,M4 Pro,48,Ollama,13,None,2026-09", "ram_gb": 48, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 13, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Ultra", "id": "llmcheck-glm-5-3-m5-ultra-q2-k-mlx", "model": "GLM-5.3", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "753B-A40B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q2_K", "quote": "GLM-5.3,753B-A40B,Q2_K,M5 Ultra,256,MLX,27,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 27, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Ultra", "id": "llmcheck-glm-5-3-m3-ultra-q2-k-mlx", "model": "GLM-5.3", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "753B-A40B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q2_K", "quote": "GLM-5.3,753B-A40B,Q2_K,M3 Ultra,256,MLX,19,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 19, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Ultra", "id": "llmcheck-glm-5-3-m5-ultra-q4-k-m-mlx", "model": "GLM-5.3", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "753B-A40B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-5.3,753B-A40B,Q4_K_M,M5 Ultra,512,MLX,17,None,2026-09", "ram_gb": 512, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Ultra", "id": "llmcheck-glm-5-3-m3-ultra-q4-k-m-mlx", "model": "GLM-5.3", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "753B-A40B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GLM-5.3,753B-A40B,Q4_K_M,M3 Ultra,512,MLX,11,None,2026-09", "ram_gb": 512, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Max", "id": "llmcheck-qwen-3-5-122b-a10b-m5-max-q4-k-m-mlx", "model": "Qwen 3.5 122B-A10B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "122B-A10B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 122B-A10B,122B-A10B,Q4_K_M,M5 Max,128,MLX,47,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 47, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M4 Max", "id": "llmcheck-qwen-3-5-122b-a10b-m4-max-q4-k-m-mlx", "model": "Qwen 3.5 122B-A10B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "122B-A10B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 122B-A10B,122B-A10B,Q4_K_M,M4 Max,128,MLX,42,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 42, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Max", "id": "llmcheck-qwen-3-5-122b-a10b-m3-max-q4-k-m-mlx", "model": "Qwen 3.5 122B-A10B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "122B-A10B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Qwen 3.5 122B-A10B,122B-A10B,Q4_K_M,M3 Max,128,MLX,32,None,2026-09", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 32, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M3 Ultra", "id": "llmcheck-mimo-v2-6-flash-m3-ultra-mxfp4-mlx", "model": "MiMo-V2.6-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "309B-A15B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "MiMo-V2.6-Flash,309B-A15B,MXFP4,M3 Ultra,256,MLX,30,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-03", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-09", "hardware": "M5 Ultra", "id": "llmcheck-mimo-v2-6-flash-m5-ultra-mxfp4-mlx", "model": "MiMo-V2.6-Flash", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "309B-A15B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "MiMo-V2.6-Flash,309B-A15B,MXFP4,M5 Ultra,256,MLX,42,None,2026-09", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 42, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-03", "hardware": "M5 Max", "id": "llmcheck-gpt-oss-120b-m5-max-mxfp4-ollama", "model": "GPT-oss 120B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "117B-A5.1B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 120B,117B-A5.1B,MXFP4,M5 Max,128,Ollama,58,None,2026-03", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 58, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M1 Ultra", "id": "llmcheck-gpt-oss-120b-m1-ultra-mxfp4-ollama", "model": "GPT-oss 120B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "117B-A5.1B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 120B,117B-A5.1B,MXFP4,M1 Ultra,128,Ollama,72,None,2026-08", "ram_gb": 128, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 72, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-08", "hardware": "M5 Ultra", "id": "llmcheck-gpt-oss-120b-m5-ultra-mxfp4-mlx", "model": "GPT-oss 120B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "117B-A5.1B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 120B,117B-A5.1B,MXFP4,M5 Ultra,96,MLX,97,None,2026-08", "ram_gb": 96, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 97, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M5 Max", "id": "llmcheck-gpt-oss-20b-m5-max-mxfp4-mlx", "model": "GPT-oss 20B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "21B-A3.6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 20B,21B-A3.6B,MXFP4,M5 Max,64,MLX,103,None,2026-10", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 103, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M4 Max", "id": "llmcheck-gpt-oss-20b-m4-max-mxfp4-mlx", "model": "GPT-oss 20B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "21B-A3.6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 20B,21B-A3.6B,MXFP4,M4 Max,64,MLX,97,None,2026-10", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 97, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M4 Pro", "id": "llmcheck-gpt-oss-20b-m4-pro-mxfp4-mlx", "model": "GPT-oss 20B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "21B-A3.6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 20B,21B-A3.6B,MXFP4,M4 Pro,24,MLX,64,None,2026-10", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 64, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M4", "id": "llmcheck-gpt-oss-20b-m4-mxfp4-mlx", "model": "GPT-oss 20B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "21B-A3.6B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "MXFP4", "quote": "GPT-oss 20B,21B-A3.6B,MXFP4,M4,16,MLX,34,None,2026-10", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 34, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M5 Max", "id": "llmcheck-gemma-4-12b-m5-max-q4-k-m-mlx", "model": "Gemma 4 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 12B,12B,Q4_K_M,M5 Max,64,MLX,64,None,2026-10", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 64, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M4 Max", "id": "llmcheck-gemma-4-12b-m4-max-q4-k-m-mlx", "model": "Gemma 4 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 12B,12B,Q4_K_M,M4 Max,64,MLX,57,None,2026-10", "ram_gb": 64, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 57, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M4 Pro", "id": "llmcheck-gemma-4-12b-m4-pro-q4-k-m-mlx", "model": "Gemma 4 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 12B,12B,Q4_K_M,M4 Pro,24,MLX,30,None,2026-10", "ram_gb": 24, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 30, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M4", "id": "llmcheck-gemma-4-12b-m4-q4-k-m-mlx", "model": "Gemma 4 12B", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "12B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Gemma 4 12B,12B,Q4_K_M,M4,16,MLX,13,None,2026-10", "ram_gb": 16, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 13, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M5 Ultra", "id": "llmcheck-minimax-m2-5-m5-ultra-q4-k-m-mlx", "model": "MiniMax M2.5", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "230B-A10B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "MiniMax M2.5,230B-A10B,Q4_K_M,M5 Ultra,256,MLX,57,None,2026-10", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 57, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": "2026-10", "hardware": "M3 Ultra", "id": "llmcheck-minimax-m2-5-m3-ultra-q4-k-m-mlx", "model": "MiniMax M2.5", "notes": "LLMCheck model-based estimate (bandwidth model)", "params": "230B-A10B", "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "MiniMax M2.5,230B-A10B,Q4_K_M,M3 Ultra,256,MLX,41,None,2026-10", "ram_gb": 256, "retrieved": "2026-10-05", "source_name": "LLMCheck Apple Silicon LLM Benchmark Database (CC BY 4.0)", "source_url": "https://llmcheck.net/data/benchmarks.json", "tg_tokens": null, "tps": 41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090 Laptop GPU", "id": "llmconfigurator-est-rtx-5090-laptop-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.72 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090 Laptop GPU,llama-3.1-8b,Q4_K_M,ctx 4096,107.61 tok/s (estimated, rtx-5090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090-laptop", "tps": 107.61, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090 Laptop GPU", "id": "llmconfigurator-est-rtx-5090-laptop-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.78 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090 Laptop GPU,qwen-2.5-7b,Q4_K_M,ctx 4096,116.56 tok/s (estimated, rtx-5090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090-laptop", "tps": 116.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090 Laptop GPU", "id": "llmconfigurator-est-rtx-5090-laptop-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.45 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090 Laptop GPU,qwen3-14b,Q4_K_M,ctx 4096,67.3 tok/s (estimated, rtx-5090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090-laptop", "tps": 67.3, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090 Laptop GPU", "id": "llmconfigurator-est-rtx-5090-laptop-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 1.2 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090 Laptop GPU,qwen3-30b-a3b,Q4_K_M,ctx 4096,185.5 tok/s (estimated, rtx-5090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090-laptop", "tps": 185.5, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090 Laptop GPU", "id": "llmconfigurator-est-rtx-5090-laptop-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.7 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090 Laptop GPU,qwen3-8b,Q4_K_M,ctx 4096,104.83 tok/s (estimated, rtx-5090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090-laptop", "tps": 104.83, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Ultra", "id": "llmconfigurator-est-apple-m5-ultra-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 120 W (package-spec); 0.76 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Ultra,llama-3.1-8b,Q4_K_M,ctx 4096,91.18 tok/s (estimated, apple-m5-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-ultra", "tps": 91.18, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Ultra", "id": "llmconfigurator-est-apple-m5-ultra-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 120 W (package-spec); 0.82 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Ultra,qwen-2.5-7b,Q4_K_M,ctx 4096,98.42 tok/s (estimated, apple-m5-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-ultra", "tps": 98.42, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Ultra", "id": "llmconfigurator-est-apple-m5-ultra-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 120 W (package-spec); 0.48 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Ultra,qwen3-14b,Q4_K_M,ctx 4096,57.93 tok/s (estimated, apple-m5-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-ultra", "tps": 57.93, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Ultra", "id": "llmconfigurator-est-apple-m5-ultra-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 120 W (package-spec); 1.3 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Ultra,qwen3-30b-a3b,Q4_K_M,ctx 4096,152.54 tok/s (estimated, apple-m5-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-ultra", "tps": 152.54, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Ultra", "id": "llmconfigurator-est-apple-m5-ultra-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 120 W (package-spec); 0.74 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Ultra,qwen3-8b,Q4_K_M,ctx 4096,88.92 tok/s (estimated, apple-m5-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-ultra", "tps": 88.92, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B70", "id": "llmconfigurator-est-arc-pro-b70-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 225 W (board-spec); 0.24 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 225, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B70,llama-3.1-8b,Q4_K_M,ctx 4096,53.44 tok/s (estimated, arc-pro-b70)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b70", "tps": 53.44, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B70", "id": "llmconfigurator-est-arc-pro-b70-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 225 W (board-spec); 0.26 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 225, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B70,qwen-2.5-7b,Q4_K_M,ctx 4096,58.07 tok/s (estimated, arc-pro-b70)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b70", "tps": 58.07, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B70", "id": "llmconfigurator-est-arc-pro-b70-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 225 W (board-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 225, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B70,qwen3-14b,Q4_K_M,ctx 4096,32.97 tok/s (estimated, arc-pro-b70)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b70", "tps": 32.97, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B70", "id": "llmconfigurator-est-arc-pro-b70-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 225 W (board-spec); 0.42 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 225, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B70,qwen3-30b-a3b,Q4_K_M,ctx 4096,94.65 tok/s (estimated, arc-pro-b70)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b70", "tps": 94.65, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B70", "id": "llmconfigurator-est-arc-pro-b70-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 225 W (board-spec); 0.23 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 225, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B70,qwen3-8b,Q4_K_M,ctx 4096,52.01 tok/s (estimated, arc-pro-b70)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b70", "tps": 52.01, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B65", "id": "llmconfigurator-est-arc-pro-b65-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 200 W (board-spec); 0.27 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B65,llama-3.1-8b,Q4_K_M,ctx 4096,53.44 tok/s (estimated, arc-pro-b65)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b65", "tps": 53.44, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B65", "id": "llmconfigurator-est-arc-pro-b65-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 200 W (board-spec); 0.29 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B65,qwen-2.5-7b,Q4_K_M,ctx 4096,58.07 tok/s (estimated, arc-pro-b65)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b65", "tps": 58.07, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B65", "id": "llmconfigurator-est-arc-pro-b65-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 200 W (board-spec); 0.16 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B65,qwen3-14b,Q4_K_M,ctx 4096,32.97 tok/s (estimated, arc-pro-b65)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b65", "tps": 32.97, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B65", "id": "llmconfigurator-est-arc-pro-b65-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 200 W (board-spec); 0.47 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B65,qwen3-30b-a3b,Q4_K_M,ctx 4096,94.65 tok/s (estimated, arc-pro-b65)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b65", "tps": 94.65, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc Pro B65", "id": "llmconfigurator-est-arc-pro-b65-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 200 W (board-spec); 0.26 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc Pro B65,qwen3-8b,Q4_K_M,ctx 4096,52.01 tok/s (estimated, arc-pro-b65)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-pro-b65", "tps": 52.01, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon AI PRO R9700", "id": "llmconfigurator-est-radeon-ai-pro-r9700-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.22 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon AI PRO R9700,llama-3.1-8b,Q4_K_M,ctx 4096,64.6 tok/s (estimated, radeon-ai-pro-r9700)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/radeon-ai-pro-r9700", "tps": 64.6, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon AI PRO R9700", "id": "llmconfigurator-est-radeon-ai-pro-r9700-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.23 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon AI PRO R9700,qwen-2.5-7b,Q4_K_M,ctx 4096,70.09 tok/s (estimated, radeon-ai-pro-r9700)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/radeon-ai-pro-r9700", "tps": 70.09, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon AI PRO R9700", "id": "llmconfigurator-est-radeon-ai-pro-r9700-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.13 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon AI PRO R9700,qwen3-14b,Q4_K_M,ctx 4096,40.09 tok/s (estimated, radeon-ai-pro-r9700)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/radeon-ai-pro-r9700", "tps": 40.09, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon AI PRO R9700", "id": "llmconfigurator-est-radeon-ai-pro-r9700-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.38 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon AI PRO R9700,qwen3-30b-a3b,Q4_K_M,ctx 4096,113.04 tok/s (estimated, radeon-ai-pro-r9700)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/radeon-ai-pro-r9700", "tps": 113.04, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon AI PRO R9700", "id": "llmconfigurator-est-radeon-ai-pro-r9700-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.21 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon AI PRO R9700,qwen3-8b,Q4_K_M,ctx 4096,62.9 tok/s (estimated, radeon-ai-pro-r9700)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/radeon-ai-pro-r9700", "tps": 62.9, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090 Ti", "id": "llmconfigurator-est-rtx-3090-ti-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 450 W (board-spec); 0.21 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090 Ti,llama-3.1-8b,Q4_K_M,ctx 4096,92.58 tok/s (estimated, rtx-3090-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090-ti", "tps": 92.58, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090 Ti", "id": "llmconfigurator-est-rtx-3090-ti-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 450 W (board-spec); 0.22 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090 Ti,qwen-2.5-7b,Q4_K_M,ctx 4096,100.65 tok/s (estimated, rtx-3090-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090-ti", "tps": 100.65, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090 Ti", "id": "llmconfigurator-est-rtx-3090-ti-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 450 W (board-spec); 0.13 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090 Ti,qwen3-14b,Q4_K_M,ctx 4096,56.95 tok/s (estimated, rtx-3090-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090-ti", "tps": 56.95, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090 Ti", "id": "llmconfigurator-est-rtx-3090-ti-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 450 W (board-spec); 0.37 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090 Ti,qwen3-30b-a3b,Q4_K_M,ctx 4096,164.9 tok/s (estimated, rtx-3090-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090-ti", "tps": 164.9, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090 Ti", "id": "llmconfigurator-est-rtx-3090-ti-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 450 W (board-spec); 0.2 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090 Ti,qwen3-8b,Q4_K_M,ctx 4096,90.08 tok/s (estimated, rtx-3090-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090-ti", "tps": 90.08, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 Ti", "id": "llmconfigurator-est-rtx-3080-ti-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.24 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 Ti,llama-3.1-8b,Q4_K_M,ctx 4096,85.45 tok/s (estimated, rtx-3080-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-ti", "tps": 85.45, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 Ti", "id": "llmconfigurator-est-rtx-3080-ti-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.27 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 Ti,qwen-2.5-7b,Q4_K_M,ctx 4096,93.07 tok/s (estimated, rtx-3080-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-ti", "tps": 93.07, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 Ti", "id": "llmconfigurator-est-rtx-3080-ti-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 Ti,qwen3-14b,Q4_K_M,ctx 4096,52.16 tok/s (estimated, rtx-3080-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-ti", "tps": 52.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 Ti", "id": "llmconfigurator-est-rtx-3080-ti-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.24 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 Ti,qwen3-8b,Q4_K_M,ctx 4096,83.1 tok/s (estimated, rtx-3080-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-ti", "tps": 83.1, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 12GB", "id": "llmconfigurator-est-rtx-3080-12-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.24 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 12GB,llama-3.1-8b,Q4_K_M,ctx 4096,85.45 tok/s (estimated, rtx-3080-12)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-12", "tps": 85.45, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 12GB", "id": "llmconfigurator-est-rtx-3080-12-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.27 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 12GB,qwen-2.5-7b,Q4_K_M,ctx 4096,93.07 tok/s (estimated, rtx-3080-12)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-12", "tps": 93.07, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 12GB", "id": "llmconfigurator-est-rtx-3080-12-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 12GB,qwen3-14b,Q4_K_M,ctx 4096,52.16 tok/s (estimated, rtx-3080-12)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-12", "tps": 52.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 12GB", "id": "llmconfigurator-est-rtx-3080-12-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.24 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 12GB,qwen3-8b,Q4_K_M,ctx 4096,83.1 tok/s (estimated, rtx-3080-12)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080-12", "tps": 83.1, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 8GB", "id": "llmconfigurator-est-rtx-4060-ti-8-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 160 W (board-spec); 0.28 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 8GB,llama-3.1-8b,Q4_K_M,ctx 4096,44.75 tok/s (estimated, rtx-4060-ti-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-8", "tps": 44.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 8GB", "id": "llmconfigurator-est-rtx-4060-ti-8-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 160 W (board-spec); 0.31 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 8GB,qwen-2.5-7b,Q4_K_M,ctx 4096,49.23 tok/s (estimated, rtx-4060-ti-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-8", "tps": 49.23, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 8GB", "id": "llmconfigurator-est-rtx-4060-ti-8-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 160 W (board-spec); 0.27 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 8GB,qwen3-8b,Q4_K_M,ctx 4096,43.38 tok/s (estimated, rtx-4060-ti-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-8", "tps": 43.38, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7700 XT", "id": "llmconfigurator-est-rx-7700-xt-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 245 W (board-spec); 0.18 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 245, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7700 XT,llama-3.1-8b,Q4_K_M,ctx 4096,44.22 tok/s (estimated, rx-7700-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7700-xt", "tps": 44.22, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7700 XT", "id": "llmconfigurator-est-rx-7700-xt-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 245 W (board-spec); 0.2 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 245, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7700 XT,qwen-2.5-7b,Q4_K_M,ctx 4096,48.36 tok/s (estimated, rx-7700-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7700-xt", "tps": 48.36, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7700 XT", "id": "llmconfigurator-est-rx-7700-xt-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 245 W (board-spec); 0.11 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 245, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7700 XT,qwen3-14b,Q4_K_M,ctx 4096,26.52 tok/s (estimated, rx-7700-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7700-xt", "tps": 26.52, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7700 XT", "id": "llmconfigurator-est-rx-7700-xt-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 245 W (board-spec); 0.18 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 245, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7700 XT,qwen3-8b,Q4_K_M,ctx 4096,42.95 tok/s (estimated, rx-7700-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7700-xt", "tps": 42.95, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090 Laptop GPU", "id": "llmconfigurator-est-rtx-4090-laptop-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 150 W (board-spec); 0.54 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090 Laptop GPU,llama-3.1-8b,Q4_K_M,ctx 4096,81.32 tok/s (estimated, rtx-4090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090-laptop", "tps": 81.32, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090 Laptop GPU", "id": "llmconfigurator-est-rtx-4090-laptop-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 150 W (board-spec); 0.59 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090 Laptop GPU,qwen-2.5-7b,Q4_K_M,ctx 4096,88.66 tok/s (estimated, rtx-4090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090-laptop", "tps": 88.66, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090 Laptop GPU", "id": "llmconfigurator-est-rtx-4090-laptop-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 150 W (board-spec); 0.33 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090 Laptop GPU,qwen3-14b,Q4_K_M,ctx 4096,49.41 tok/s (estimated, rtx-4090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090-laptop", "tps": 49.41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090 Laptop GPU", "id": "llmconfigurator-est-rtx-4090-laptop-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 150 W (board-spec); 0.53 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090 Laptop GPU,qwen3-8b,Q4_K_M,ctx 4096,79.06 tok/s (estimated, rtx-4090-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090-laptop", "tps": 79.06, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Laptop GPU", "id": "llmconfigurator-est-rtx-4060-laptop-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 115 W (board-spec); 0.35 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 115, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Laptop GPU,llama-3.1-8b,Q4_K_M,ctx 4096,40.22 tok/s (estimated, rtx-4060-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-laptop", "tps": 40.22, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Laptop GPU", "id": "llmconfigurator-est-rtx-4060-laptop-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 115 W (board-spec); 0.39 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 115, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Laptop GPU,qwen-2.5-7b,Q4_K_M,ctx 4096,44.3 tok/s (estimated, rtx-4060-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-laptop", "tps": 44.3, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Laptop GPU", "id": "llmconfigurator-est-rtx-4060-laptop-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 115 W (board-spec); 0.34 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 115, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Laptop GPU,qwen3-8b,Q4_K_M,ctx 4096,38.98 tok/s (estimated, rtx-4060-laptop)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-laptop", "tps": 38.98, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI 9 HX 370", "id": "llmconfigurator-est-amd-ryzen-ai-9-hx-370-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 54 W (package-spec); 0.26 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 54, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI 9 HX 370,llama-3.1-8b,Q4_K_M,ctx 4096,13.83 tok/s (estimated, amd-ryzen-ai-9-hx-370)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-9-hx-370", "tps": 13.83, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI 9 HX 370", "id": "llmconfigurator-est-amd-ryzen-ai-9-hx-370-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 54 W (package-spec); 0.28 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 54, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI 9 HX 370,qwen-2.5-7b,Q4_K_M,ctx 4096,15.3 tok/s (estimated, amd-ryzen-ai-9-hx-370)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-9-hx-370", "tps": 15.3, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI 9 HX 370", "id": "llmconfigurator-est-amd-ryzen-ai-9-hx-370-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 54 W (package-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 54, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI 9 HX 370,qwen3-14b,Q4_K_M,ctx 4096,7.9 tok/s (estimated, amd-ryzen-ai-9-hx-370)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-9-hx-370", "tps": 7.9, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI 9 HX 370", "id": "llmconfigurator-est-amd-ryzen-ai-9-hx-370-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 54 W (package-spec); 0.54 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 54, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI 9 HX 370,qwen3-30b-a3b,Q4_K_M,ctx 4096,29.24 tok/s (estimated, amd-ryzen-ai-9-hx-370)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-9-hx-370", "tps": 29.24, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI 9 HX 370", "id": "llmconfigurator-est-amd-ryzen-ai-9-hx-370-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 54 W (package-spec); 0.25 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 54, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI 9 HX 370,qwen3-8b,Q4_K_M,ctx 4096,13.38 tok/s (estimated, amd-ryzen-ai-9-hx-370)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-9-hx-370", "tps": 13.38, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "GB10 Grace Blackwell", "id": "llmconfigurator-est-nvidia-gb10-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 140 W (package-spec); 0.28 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 140, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GB10 Grace Blackwell,llama-3.1-8b,Q4_K_M,ctx 4096,39.41 tok/s (estimated, nvidia-gb10)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-gb10", "tps": 39.41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "GB10 Grace Blackwell", "id": "llmconfigurator-est-nvidia-gb10-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 140 W (package-spec); 0.31 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 140, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GB10 Grace Blackwell,qwen-2.5-7b,Q4_K_M,ctx 4096,43.42 tok/s (estimated, nvidia-gb10)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-gb10", "tps": 43.42, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "GB10 Grace Blackwell", "id": "llmconfigurator-est-nvidia-gb10-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 140 W (package-spec); 0.16 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 140, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GB10 Grace Blackwell,qwen3-14b,Q4_K_M,ctx 4096,22.91 tok/s (estimated, nvidia-gb10)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-gb10", "tps": 22.91, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "GB10 Grace Blackwell", "id": "llmconfigurator-est-nvidia-gb10-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 140 W (package-spec); 0.57 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 140, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GB10 Grace Blackwell,qwen3-30b-a3b,Q4_K_M,ctx 4096,79.57 tok/s (estimated, nvidia-gb10)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-gb10", "tps": 79.57, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "GB10 Grace Blackwell", "id": "llmconfigurator-est-nvidia-gb10-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 140 W (package-spec); 0.27 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 140, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "GB10 Grace Blackwell,qwen3-8b,Q4_K_M,ctx 4096,38.19 tok/s (estimated, nvidia-gb10)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-gb10", "tps": 38.19, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max (30-core GPU)", "id": "llmconfigurator-est-apple-m3-max-30c-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.8 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max (30-core GPU),llama-3.1-8b,Q4_K_M,ctx 4096,28.06 tok/s (estimated, apple-m3-max-30c)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max-30c", "tps": 28.06, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max (30-core GPU)", "id": "llmconfigurator-est-apple-m3-max-30c-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.88 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max (30-core GPU),qwen-2.5-7b,Q4_K_M,ctx 4096,30.93 tok/s (estimated, apple-m3-max-30c)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max-30c", "tps": 30.93, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max (30-core GPU)", "id": "llmconfigurator-est-apple-m3-max-30c-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.47 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max (30-core GPU),qwen3-14b,Q4_K_M,ctx 4096,16.28 tok/s (estimated, apple-m3-max-30c)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max-30c", "tps": 16.28, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max (30-core GPU)", "id": "llmconfigurator-est-apple-m3-max-30c-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.6 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max (30-core GPU),qwen3-30b-a3b,Q4_K_M,ctx 4096,56.93 tok/s (estimated, apple-m3-max-30c)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max-30c", "tps": 56.93, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max (30-core GPU)", "id": "llmconfigurator-est-apple-m3-max-30c-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.78 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max (30-core GPU),qwen3-8b,Q4_K_M,ctx 4096,27.19 tok/s (estimated, apple-m3-max-30c)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max-30c", "tps": 27.19, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090", "id": "llmconfigurator-est-rtx-5090-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 575 W (board-spec); 0.3 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 575, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090,llama-3.1-8b,Q4_K_M,ctx 4096,173.33 tok/s (estimated, rtx-5090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090", "tps": 173.33, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090", "id": "llmconfigurator-est-rtx-5090-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 575 W (board-spec); 0.32 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 575, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090,qwen-2.5-7b,Q4_K_M,ctx 4096,184.75 tok/s (estimated, rtx-5090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090", "tps": 184.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090", "id": "llmconfigurator-est-rtx-5090-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 575 W (board-spec); 0.2 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 575, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090,qwen3-14b,Q4_K_M,ctx 4096,116.92 tok/s (estimated, rtx-5090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090", "tps": 116.92, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090", "id": "llmconfigurator-est-rtx-5090-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 575 W (board-spec); 0.46 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 575, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090,qwen3-30b-a3b,Q4_K_M,ctx 4096,261.89 tok/s (estimated, rtx-5090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090", "tps": 261.89, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5090", "id": "llmconfigurator-est-rtx-5090-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 575 W (board-spec); 0.3 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 575, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5090,qwen3-8b,Q4_K_M,ctx 4096,169.7 tok/s (estimated, rtx-5090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5090", "tps": 169.7, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5080", "id": "llmconfigurator-est-rtx-5080-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 360 W (board-spec); 0.31 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 360, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5080,llama-3.1-8b,Q4_K_M,ctx 4096,113.34 tok/s (estimated, rtx-5080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5080", "tps": 113.34, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5080", "id": "llmconfigurator-est-rtx-5080-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 360 W (board-spec); 0.34 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 360, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5080,qwen-2.5-7b,Q4_K_M,ctx 4096,122.59 tok/s (estimated, rtx-5080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5080", "tps": 122.59, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5080", "id": "llmconfigurator-est-rtx-5080-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 360 W (board-spec); 0.2 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 360, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5080,qwen3-14b,Q4_K_M,ctx 4096,71.33 tok/s (estimated, rtx-5080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5080", "tps": 71.33, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5080", "id": "llmconfigurator-est-rtx-5080-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 360 W (board-spec); 0.31 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 360, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5080,qwen3-8b,Q4_K_M,ctx 4096,110.46 tok/s (estimated, rtx-5080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5080", "tps": 110.46, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070 Ti", "id": "llmconfigurator-est-rtx-5070-ti-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.36 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070 Ti,llama-3.1-8b,Q4_K_M,ctx 4096,107.61 tok/s (estimated, rtx-5070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070-ti", "tps": 107.61, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070 Ti", "id": "llmconfigurator-est-rtx-5070-ti-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.39 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070 Ti,qwen-2.5-7b,Q4_K_M,ctx 4096,116.56 tok/s (estimated, rtx-5070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070-ti", "tps": 116.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070 Ti", "id": "llmconfigurator-est-rtx-5070-ti-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.22 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070 Ti,qwen3-14b,Q4_K_M,ctx 4096,67.3 tok/s (estimated, rtx-5070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070-ti", "tps": 67.3, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070 Ti", "id": "llmconfigurator-est-rtx-5070-ti-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 300 W (board-spec); 0.35 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070 Ti,qwen3-8b,Q4_K_M,ctx 4096,104.83 tok/s (estimated, rtx-5070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070-ti", "tps": 104.83, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070", "id": "llmconfigurator-est-rtx-5070-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 250 W (board-spec); 0.34 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 250, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070,llama-3.1-8b,Q4_K_M,ctx 4096,85.9 tok/s (estimated, rtx-5070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070", "tps": 85.9, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070", "id": "llmconfigurator-est-rtx-5070-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 250 W (board-spec); 0.37 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 250, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070,qwen-2.5-7b,Q4_K_M,ctx 4096,93.54 tok/s (estimated, rtx-5070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070", "tps": 93.54, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070", "id": "llmconfigurator-est-rtx-5070-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 250 W (board-spec); 0.21 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 250, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070,qwen3-14b,Q4_K_M,ctx 4096,52.45 tok/s (estimated, rtx-5070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070", "tps": 52.45, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5070", "id": "llmconfigurator-est-rtx-5070-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 250 W (board-spec); 0.33 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 250, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5070,qwen3-8b,Q4_K_M,ctx 4096,83.54 tok/s (estimated, rtx-5070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5070", "tps": 83.54, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 16GB", "id": "llmconfigurator-est-rtx-5060-ti-16-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.34 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 16GB,llama-3.1-8b,Q4_K_M,ctx 4096,61.2 tok/s (estimated, rtx-5060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-16", "tps": 61.2, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 16GB", "id": "llmconfigurator-est-rtx-5060-ti-16-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.37 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 16GB,qwen-2.5-7b,Q4_K_M,ctx 4096,67.06 tok/s (estimated, rtx-5060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-16", "tps": 67.06, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 16GB", "id": "llmconfigurator-est-rtx-5060-ti-16-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.2 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 16GB,qwen3-14b,Q4_K_M,ctx 4096,36.4 tok/s (estimated, rtx-5060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-16", "tps": 36.4, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 16GB", "id": "llmconfigurator-est-rtx-5060-ti-16-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.33 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 16GB,qwen3-8b,Q4_K_M,ctx 4096,59.41 tok/s (estimated, rtx-5060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-16", "tps": 59.41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 8GB", "id": "llmconfigurator-est-rtx-5060-ti-8-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.34 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 8GB,llama-3.1-8b,Q4_K_M,ctx 4096,61.2 tok/s (estimated, rtx-5060-ti-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-8", "tps": 61.2, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 8GB", "id": "llmconfigurator-est-rtx-5060-ti-8-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.37 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 8GB,qwen-2.5-7b,Q4_K_M,ctx 4096,67.06 tok/s (estimated, rtx-5060-ti-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-8", "tps": 67.06, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060 Ti 8GB", "id": "llmconfigurator-est-rtx-5060-ti-8-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 180 W (board-spec); 0.33 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 180, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060 Ti 8GB,qwen3-8b,Q4_K_M,ctx 4096,59.41 tok/s (estimated, rtx-5060-ti-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060-ti-8", "tps": 59.41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060", "id": "llmconfigurator-est-rtx-5060-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 145 W (board-spec); 0.42 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 145, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060,llama-3.1-8b,Q4_K_M,ctx 4096,61.2 tok/s (estimated, rtx-5060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060", "tps": 61.2, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060", "id": "llmconfigurator-est-rtx-5060-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 145 W (board-spec); 0.46 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 145, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060,qwen-2.5-7b,Q4_K_M,ctx 4096,67.06 tok/s (estimated, rtx-5060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060", "tps": 67.06, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 5060", "id": "llmconfigurator-est-rtx-5060-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 145 W (board-spec); 0.41 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 145, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 5060,qwen3-8b,Q4_K_M,ctx 4096,59.41 tok/s (estimated, rtx-5060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-5060", "tps": 59.41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti", "id": "llmconfigurator-est-rtx-4070-ti-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.26 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti,llama-3.1-8b,Q4_K_M,ctx 4096,72.82 tok/s (estimated, rtx-4070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti", "tps": 72.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti", "id": "llmconfigurator-est-rtx-4070-ti-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.28 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti,qwen-2.5-7b,Q4_K_M,ctx 4096,79.56 tok/s (estimated, rtx-4070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti", "tps": 79.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti", "id": "llmconfigurator-est-rtx-4070-ti-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti,qwen3-14b,Q4_K_M,ctx 4096,43.84 tok/s (estimated, rtx-4070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti", "tps": 43.84, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti", "id": "llmconfigurator-est-rtx-4070-ti-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.25 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti,qwen3-8b,Q4_K_M,ctx 4096,70.75 tok/s (estimated, rtx-4070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti", "tps": 70.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 SUPER", "id": "llmconfigurator-est-rtx-4070-super-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 220 W (board-spec); 0.33 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Super,llama-3.1-8b,Q4_K_M,ctx 4096,72.82 tok/s (estimated, rtx-4070-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-super", "tps": 72.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 SUPER", "id": "llmconfigurator-est-rtx-4070-super-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 220 W (board-spec); 0.36 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Super,qwen-2.5-7b,Q4_K_M,ctx 4096,79.56 tok/s (estimated, rtx-4070-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-super", "tps": 79.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 SUPER", "id": "llmconfigurator-est-rtx-4070-super-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 220 W (board-spec); 0.2 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Super,qwen3-14b,Q4_K_M,ctx 4096,43.84 tok/s (estimated, rtx-4070-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-super", "tps": 43.84, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 SUPER", "id": "llmconfigurator-est-rtx-4070-super-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 220 W (board-spec); 0.32 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Super,qwen3-8b,Q4_K_M,ctx 4096,70.75 tok/s (estimated, rtx-4070-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-super", "tps": 70.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090", "id": "llmconfigurator-est-rtx-4090-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 450 W (board-spec); 0.28 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090,llama-3.1-8b,Q4_K_M,ctx 4096,125.16 tok/s (estimated, rtx-4090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090", "tps": 125.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090", "id": "llmconfigurator-est-rtx-4090-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 450 W (board-spec); 0.3 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090,qwen-2.5-7b,Q4_K_M,ctx 4096,135 tok/s (estimated, rtx-4090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090", "tps": 135, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090", "id": "llmconfigurator-est-rtx-4090-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 450 W (board-spec); 0.18 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090,qwen3-14b,Q4_K_M,ctx 4096,79.83 tok/s (estimated, rtx-4090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090", "tps": 79.83, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090", "id": "llmconfigurator-est-rtx-4090-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 450 W (board-spec); 0.46 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090,qwen3-30b-a3b,Q4_K_M,ctx 4096,207.94 tok/s (estimated, rtx-4090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090", "tps": 207.94, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4090", "id": "llmconfigurator-est-rtx-4090-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 450 W (board-spec); 0.31 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 450, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4090,qwen3-8b,Q4_K_M,ctx 4096,141.3 tok/s (estimated, rtx-4090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4090", "tps": 141.3, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080 SUPER", "id": "llmconfigurator-est-rtx-4080-super-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.31 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080 Super,llama-3.1-8b,Q4_K_M,ctx 4096,98.89 tok/s (estimated, rtx-4080-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080-super", "tps": 98.89, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080 SUPER", "id": "llmconfigurator-est-rtx-4080-super-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.34 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080 Super,qwen-2.5-7b,Q4_K_M,ctx 4096,107.35 tok/s (estimated, rtx-4080-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080-super", "tps": 107.35, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080 SUPER", "id": "llmconfigurator-est-rtx-4080-super-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.19 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080 Super,qwen3-14b,Q4_K_M,ctx 4096,61.25 tok/s (estimated, rtx-4080-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080-super", "tps": 61.25, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080 SUPER", "id": "llmconfigurator-est-rtx-4080-super-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.3 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080 Super,qwen3-8b,Q4_K_M,ctx 4096,96.27 tok/s (estimated, rtx-4080-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080-super", "tps": 96.27, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080", "id": "llmconfigurator-est-rtx-4080-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.3 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080,llama-3.1-8b,Q4_K_M,ctx 4096,96.89 tok/s (estimated, rtx-4080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080", "tps": 96.89, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080", "id": "llmconfigurator-est-rtx-4080-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.33 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080,qwen-2.5-7b,Q4_K_M,ctx 4096,105.23 tok/s (estimated, rtx-4080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080", "tps": 105.23, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080", "id": "llmconfigurator-est-rtx-4080-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.19 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080,qwen3-14b,Q4_K_M,ctx 4096,59.88 tok/s (estimated, rtx-4080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080", "tps": 59.88, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4080", "id": "llmconfigurator-est-rtx-4080-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 320 W (board-spec); 0.29 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4080,qwen3-8b,Q4_K_M,ctx 4096,94.31 tok/s (estimated, rtx-4080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4080", "tps": 94.31, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti SUPER", "id": "llmconfigurator-est-rtx-4070-ti-super-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.32 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti Super,llama-3.1-8b,Q4_K_M,ctx 4096,92.07 tok/s (estimated, rtx-4070-ti-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti-super", "tps": 92.07, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti SUPER", "id": "llmconfigurator-est-rtx-4070-ti-super-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.35 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti Super,qwen-2.5-7b,Q4_K_M,ctx 4096,100.11 tok/s (estimated, rtx-4070-ti-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti-super", "tps": 100.11, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti SUPER", "id": "llmconfigurator-est-rtx-4070-ti-super-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.2 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti Super,qwen3-14b,Q4_K_M,ctx 4096,56.6 tok/s (estimated, rtx-4070-ti-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti-super", "tps": 56.6, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070 Ti SUPER", "id": "llmconfigurator-est-rtx-4070-ti-super-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 285 W (board-spec); 0.31 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 285, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070 Ti Super,qwen3-8b,Q4_K_M,ctx 4096,89.58 tok/s (estimated, rtx-4070-ti-super)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070-ti-super", "tps": 89.58, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070", "id": "llmconfigurator-est-rtx-4070-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 200 W (board-spec); 0.36 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070,llama-3.1-8b,Q4_K_M,ctx 4096,72.82 tok/s (estimated, rtx-4070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070", "tps": 72.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070", "id": "llmconfigurator-est-rtx-4070-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 200 W (board-spec); 0.4 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070,qwen-2.5-7b,Q4_K_M,ctx 4096,79.56 tok/s (estimated, rtx-4070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070", "tps": 79.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070", "id": "llmconfigurator-est-rtx-4070-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 200 W (board-spec); 0.22 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070,qwen3-14b,Q4_K_M,ctx 4096,43.84 tok/s (estimated, rtx-4070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070", "tps": 43.84, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4070", "id": "llmconfigurator-est-rtx-4070-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 200 W (board-spec); 0.35 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 200, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4070,qwen3-8b,Q4_K_M,ctx 4096,70.75 tok/s (estimated, rtx-4070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4070", "tps": 70.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 16GB", "id": "llmconfigurator-est-rtx-4060-ti-16-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 165 W (board-spec); 0.27 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 165, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 16GB,llama-3.1-8b,Q4_K_M,ctx 4096,44.75 tok/s (estimated, rtx-4060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-16", "tps": 44.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 16GB", "id": "llmconfigurator-est-rtx-4060-ti-16-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 165 W (board-spec); 0.3 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 165, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 16GB,qwen-2.5-7b,Q4_K_M,ctx 4096,49.23 tok/s (estimated, rtx-4060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-16", "tps": 49.23, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 16GB", "id": "llmconfigurator-est-rtx-4060-ti-16-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 165 W (board-spec); 0.16 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 165, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 16GB,qwen3-14b,Q4_K_M,ctx 4096,26.16 tok/s (estimated, rtx-4060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-16", "tps": 26.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060 Ti 16GB", "id": "llmconfigurator-est-rtx-4060-ti-16-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 165 W (board-spec); 0.26 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 165, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060 Ti 16GB,qwen3-8b,Q4_K_M,ctx 4096,43.38 tok/s (estimated, rtx-4060-ti-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060-ti-16", "tps": 43.38, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060", "id": "llmconfigurator-est-rtx-4060-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 115 W (board-spec); 0.37 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 115, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060,llama-3.1-8b,Q4_K_M,ctx 4096,42.5 tok/s (estimated, rtx-4060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060", "tps": 42.5, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060", "id": "llmconfigurator-est-rtx-4060-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 115 W (board-spec); 0.41 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 115, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060,qwen-2.5-7b,Q4_K_M,ctx 4096,46.78 tok/s (estimated, rtx-4060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060", "tps": 46.78, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 4060", "id": "llmconfigurator-est-rtx-4060-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 115 W (board-spec); 0.36 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 115, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 4060,qwen3-8b,Q4_K_M,ctx 4096,41.19 tok/s (estimated, rtx-4060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-4060", "tps": 41.19, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3070", "id": "llmconfigurator-est-rtx-3070-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 220 W (board-spec); 0.21 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3070,llama-3.1-8b,Q4_K_M,ctx 4096,46.52 tok/s (estimated, rtx-3070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3070", "tps": 46.52, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3070", "id": "llmconfigurator-est-rtx-3070-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 220 W (board-spec); 0.23 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3070,qwen-2.5-7b,Q4_K_M,ctx 4096,51.16 tok/s (estimated, rtx-3070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3070", "tps": 51.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3070", "id": "llmconfigurator-est-rtx-3070-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 220 W (board-spec); 0.21 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3070,qwen3-8b,Q4_K_M,ctx 4096,45.11 tok/s (estimated, rtx-3070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3070", "tps": 45.11, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3070 Ti", "id": "llmconfigurator-est-rtx-3070-ti-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 290 W (board-spec); 0.21 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 290, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3070 Ti,llama-3.1-8b,Q4_K_M,ctx 4096,60.86 tok/s (estimated, rtx-3070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3070-ti", "tps": 60.86, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3070 Ti", "id": "llmconfigurator-est-rtx-3070-ti-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 290 W (board-spec); 0.23 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 290, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3070 Ti,qwen-2.5-7b,Q4_K_M,ctx 4096,66.69 tok/s (estimated, rtx-3070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3070-ti", "tps": 66.69, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3070 Ti", "id": "llmconfigurator-est-rtx-3070-ti-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 290 W (board-spec); 0.2 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 290, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3070 Ti,qwen3-8b,Q4_K_M,ctx 4096,59.08 tok/s (estimated, rtx-3070-ti)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3070-ti", "tps": 59.08, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090", "id": "llmconfigurator-est-rtx-3090-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.25 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090,llama-3.1-8b,Q4_K_M,ctx 4096,87.26 tok/s (estimated, rtx-3090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090", "tps": 87.26, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090", "id": "llmconfigurator-est-rtx-3090-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 350 W (board-spec); 0.27 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090,qwen-2.5-7b,Q4_K_M,ctx 4096,95 tok/s (estimated, rtx-3090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090", "tps": 95, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090", "id": "llmconfigurator-est-rtx-3090-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090,qwen3-14b,Q4_K_M,ctx 4096,53.37 tok/s (estimated, rtx-3090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090", "tps": 53.37, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090", "id": "llmconfigurator-est-rtx-3090-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.45 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090,qwen3-30b-a3b,Q4_K_M,ctx 4096,157.29 tok/s (estimated, rtx-3090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090", "tps": 157.29, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3090", "id": "llmconfigurator-est-rtx-3090-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 350 W (board-spec); 0.24 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3090,qwen3-8b,Q4_K_M,ctx 4096,84.87 tok/s (estimated, rtx-3090)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3090", "tps": 84.87, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 (10GB)", "id": "llmconfigurator-est-rtx-3080-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 320 W (board-spec); 0.23 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 (10GB),llama-3.1-8b,Q4_K_M,ctx 4096,73.56 tok/s (estimated, rtx-3080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080", "tps": 73.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 (10GB)", "id": "llmconfigurator-est-rtx-3080-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 320 W (board-spec); 0.25 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 (10GB),qwen-2.5-7b,Q4_K_M,ctx 4096,80.36 tok/s (estimated, rtx-3080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080", "tps": 80.36, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3080 (10GB)", "id": "llmconfigurator-est-rtx-3080-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 320 W (board-spec); 0.22 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 320, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3080 (10GB),qwen3-8b,Q4_K_M,ctx 4096,71.47 tok/s (estimated, rtx-3080)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3080", "tps": 71.47, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3060 (12GB)", "id": "llmconfigurator-est-rtx-3060-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 170 W (board-spec); 0.22 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 170, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3060 (12GB),llama-3.1-8b,Q4_K_M,ctx 4096,38.16 tok/s (estimated, rtx-3060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3060", "tps": 38.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3060 (12GB)", "id": "llmconfigurator-est-rtx-3060-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 170 W (board-spec); 0.25 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 170, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3060 (12GB),qwen-2.5-7b,Q4_K_M,ctx 4096,42.06 tok/s (estimated, rtx-3060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3060", "tps": 42.06, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3060 (12GB)", "id": "llmconfigurator-est-rtx-3060-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 170 W (board-spec); 0.13 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 170, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3060 (12GB),qwen3-14b,Q4_K_M,ctx 4096,22.16 tok/s (estimated, rtx-3060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3060", "tps": 22.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 3060 (12GB)", "id": "llmconfigurator-est-rtx-3060-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 170 W (board-spec); 0.22 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 170, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 3060 (12GB),qwen3-8b,Q4_K_M,ctx 4096,36.98 tok/s (estimated, rtx-3060)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-3060", "tps": 36.98, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XTX", "id": "llmconfigurator-est-rx-7900-xtx-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 355 W (board-spec); 0.23 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 355, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XTX,llama-3.1-8b,Q4_K_M,ctx 4096,82.64 tok/s (estimated, rx-7900-xtx)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xtx", "tps": 82.64, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XTX", "id": "llmconfigurator-est-rx-7900-xtx-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 355 W (board-spec); 0.25 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 355, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XTX,qwen-2.5-7b,Q4_K_M,ctx 4096,89.05 tok/s (estimated, rx-7900-xtx)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xtx", "tps": 89.05, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XTX", "id": "llmconfigurator-est-rx-7900-xtx-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 355 W (board-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 355, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XTX,qwen3-14b,Q4_K_M,ctx 4096,52.92 tok/s (estimated, rx-7900-xtx)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xtx", "tps": 52.92, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XTX", "id": "llmconfigurator-est-rx-7900-xtx-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 355 W (board-spec); 0.38 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 355, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XTX,qwen3-30b-a3b,Q4_K_M,ctx 4096,136.27 tok/s (estimated, rx-7900-xtx)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xtx", "tps": 136.27, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XTX", "id": "llmconfigurator-est-rx-7900-xtx-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 355 W (board-spec); 0.23 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 355, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XTX,qwen3-8b,Q4_K_M,ctx 4096,80.63 tok/s (estimated, rx-7900-xtx)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xtx", "tps": 80.63, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XT", "id": "llmconfigurator-est-rx-7900-xt-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 315 W (board-spec); 0.23 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 315, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XT,llama-3.1-8b,Q4_K_M,ctx 4096,72.35 tok/s (estimated, rx-7900-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xt", "tps": 72.35, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XT", "id": "llmconfigurator-est-rx-7900-xt-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 315 W (board-spec); 0.25 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 315, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XT,qwen-2.5-7b,Q4_K_M,ctx 4096,78.27 tok/s (estimated, rx-7900-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xt", "tps": 78.27, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XT", "id": "llmconfigurator-est-rx-7900-xt-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 315 W (board-spec); 0.14 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 315, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XT,qwen3-14b,Q4_K_M,ctx 4096,45.51 tok/s (estimated, rx-7900-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xt", "tps": 45.51, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XT", "id": "llmconfigurator-est-rx-7900-xt-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 315 W (board-spec); 0.39 tok/W; VRAM fit tight (19.62 GB required)", "params": null, "power_w": 315, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XT,qwen3-30b-a3b,Q4_K_M,ctx 4096,123.37 tok/s (estimated, rx-7900-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xt", "tps": 123.37, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7900 XT", "id": "llmconfigurator-est-rx-7900-xt-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 315 W (board-spec); 0.22 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 315, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7900 XT,qwen3-8b,Q4_K_M,ctx 4096,70.51 tok/s (estimated, rx-7900-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7900-xt", "tps": 70.51, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7800 XT", "id": "llmconfigurator-est-rx-7800-xt-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 263 W (board-spec); 0.23 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 263, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7800 XT,llama-3.1-8b,Q4_K_M,ctx 4096,59.77 tok/s (estimated, rx-7800-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7800-xt", "tps": 59.77, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7800 XT", "id": "llmconfigurator-est-rx-7800-xt-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 263 W (board-spec); 0.25 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 263, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7800 XT,qwen-2.5-7b,Q4_K_M,ctx 4096,64.97 tok/s (estimated, rx-7800-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7800-xt", "tps": 64.97, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7800 XT", "id": "llmconfigurator-est-rx-7800-xt-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 263 W (board-spec); 0.14 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 263, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7800 XT,qwen3-14b,Q4_K_M,ctx 4096,36.79 tok/s (estimated, rx-7800-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7800-xt", "tps": 36.79, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 7800 XT", "id": "llmconfigurator-est-rx-7800-xt-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 263 W (board-spec); 0.22 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 263, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 7800 XT,qwen3-8b,Q4_K_M,ctx 4096,58.16 tok/s (estimated, rx-7800-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-7800-xt", "tps": 58.16, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070 XT", "id": "llmconfigurator-est-rx-9070-xt-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 304 W (board-spec); 0.21 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 304, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070 XT,llama-3.1-8b,Q4_K_M,ctx 4096,64.6 tok/s (estimated, rx-9070-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070-xt", "tps": 64.6, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070 XT", "id": "llmconfigurator-est-rx-9070-xt-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 304 W (board-spec); 0.23 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 304, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070 XT,qwen-2.5-7b,Q4_K_M,ctx 4096,70.09 tok/s (estimated, rx-9070-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070-xt", "tps": 70.09, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070 XT", "id": "llmconfigurator-est-rx-9070-xt-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 304 W (board-spec); 0.13 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 304, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070 XT,qwen3-14b,Q4_K_M,ctx 4096,40.09 tok/s (estimated, rx-9070-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070-xt", "tps": 40.09, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070 XT", "id": "llmconfigurator-est-rx-9070-xt-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 304 W (board-spec); 0.21 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 304, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070 XT,qwen3-8b,Q4_K_M,ctx 4096,62.9 tok/s (estimated, rx-9070-xt)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070-xt", "tps": 62.9, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070", "id": "llmconfigurator-est-rx-9070-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 220 W (board-spec); 0.29 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070,llama-3.1-8b,Q4_K_M,ctx 4096,64.6 tok/s (estimated, rx-9070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070", "tps": 64.6, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070", "id": "llmconfigurator-est-rx-9070-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 220 W (board-spec); 0.32 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070,qwen-2.5-7b,Q4_K_M,ctx 4096,70.09 tok/s (estimated, rx-9070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070", "tps": 70.09, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070", "id": "llmconfigurator-est-rx-9070-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 220 W (board-spec); 0.18 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070,qwen3-14b,Q4_K_M,ctx 4096,40.09 tok/s (estimated, rx-9070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070", "tps": 40.09, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9070", "id": "llmconfigurator-est-rx-9070-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 220 W (board-spec); 0.29 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 220, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9070,qwen3-8b,Q4_K_M,ctx 4096,62.9 tok/s (estimated, rx-9070)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9070", "tps": 62.9, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 16GB", "id": "llmconfigurator-est-rx-9060-xt-16-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 160 W (board-spec); 0.23 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 16GB,llama-3.1-8b,Q4_K_M,ctx 4096,36.42 tok/s (estimated, rx-9060-xt-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-16", "tps": 36.42, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 16GB", "id": "llmconfigurator-est-rx-9060-xt-16-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 160 W (board-spec); 0.25 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 16GB,qwen-2.5-7b,Q4_K_M,ctx 4096,39.95 tok/s (estimated, rx-9060-xt-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-16", "tps": 39.95, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 16GB", "id": "llmconfigurator-est-rx-9060-xt-16-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 160 W (board-spec); 0.13 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 16GB,qwen3-14b,Q4_K_M,ctx 4096,21.56 tok/s (estimated, rx-9060-xt-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-16", "tps": 21.56, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 16GB", "id": "llmconfigurator-est-rx-9060-xt-16-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 160 W (board-spec); 0.22 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 160, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 16GB,qwen3-8b,Q4_K_M,ctx 4096,35.34 tok/s (estimated, rx-9060-xt-16)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-16", "tps": 35.34, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 8GB", "id": "llmconfigurator-est-rx-9060-xt-8-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.24 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 8GB,llama-3.1-8b,Q4_K_M,ctx 4096,36.42 tok/s (estimated, rx-9060-xt-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-8", "tps": 36.42, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 8GB", "id": "llmconfigurator-est-rx-9060-xt-8-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.27 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 8GB,qwen-2.5-7b,Q4_K_M,ctx 4096,39.95 tok/s (estimated, rx-9060-xt-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-8", "tps": 39.95, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Radeon RX 9060 XT 8GB", "id": "llmconfigurator-est-rx-9060-xt-8-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.24 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Radeon RX 9060 XT 8GB,qwen3-8b,Q4_K_M,ctx 4096,35.34 tok/s (estimated, rx-9060-xt-8)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rx-9060-xt-8", "tps": 35.34, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Max", "id": "llmconfigurator-est-apple-m1-max-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.2 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Max,llama-3.1-8b,Q4_K_M,ctx 4096,36.39 tok/s (estimated, apple-m1-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-max", "tps": 36.39, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Max", "id": "llmconfigurator-est-apple-m1-max-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.3 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Max,qwen-2.5-7b,Q4_K_M,ctx 4096,40 tok/s (estimated, apple-m1-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-max", "tps": 40, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Max", "id": "llmconfigurator-est-apple-m1-max-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.71 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Max,qwen3-14b,Q4_K_M,ctx 4096,21.36 tok/s (estimated, apple-m1-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-max", "tps": 21.36, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Max", "id": "llmconfigurator-est-apple-m1-max-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 2.4 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Max,qwen3-30b-a3b,Q4_K_M,ctx 4096,71.82 tok/s (estimated, apple-m1-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-max", "tps": 71.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Max", "id": "llmconfigurator-est-apple-m1-max-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.2 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Max,qwen3-8b,Q4_K_M,ctx 4096,35.29 tok/s (estimated, apple-m1-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-max", "tps": 35.29, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Pro", "id": "llmconfigurator-est-apple-m2-pro-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.64 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Pro,llama-3.1-8b,Q4_K_M,ctx 4096,19.25 tok/s (estimated, apple-m2-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-pro", "tps": 19.25, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Pro", "id": "llmconfigurator-est-apple-m2-pro-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.71 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Pro,qwen-2.5-7b,Q4_K_M,ctx 4096,21.28 tok/s (estimated, apple-m2-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-pro", "tps": 21.28, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Pro", "id": "llmconfigurator-est-apple-m2-pro-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.37 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Pro,qwen3-14b,Q4_K_M,ctx 4096,11.03 tok/s (estimated, apple-m2-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-pro", "tps": 11.03, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Pro", "id": "llmconfigurator-est-apple-m2-pro-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.3 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Pro,qwen3-30b-a3b,Q4_K_M,ctx 4096,40.24 tok/s (estimated, apple-m2-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-pro", "tps": 40.24, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Pro", "id": "llmconfigurator-est-apple-m2-pro-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.62 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Pro,qwen3-8b,Q4_K_M,ctx 4096,18.63 tok/s (estimated, apple-m2-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-pro", "tps": 18.63, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Max", "id": "llmconfigurator-est-apple-m2-max-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 40 W (package-spec); 0.91 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 40, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Max,llama-3.1-8b,Q4_K_M,ctx 4096,36.39 tok/s (estimated, apple-m2-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-max", "tps": 36.39, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Max", "id": "llmconfigurator-est-apple-m2-max-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 40 W (package-spec); 1 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 40, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Max,qwen-2.5-7b,Q4_K_M,ctx 4096,40 tok/s (estimated, apple-m2-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-max", "tps": 40, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Max", "id": "llmconfigurator-est-apple-m2-max-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 40 W (package-spec); 0.53 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 40, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Max,qwen3-14b,Q4_K_M,ctx 4096,21.36 tok/s (estimated, apple-m2-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-max", "tps": 21.36, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Max", "id": "llmconfigurator-est-apple-m2-max-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 40 W (package-spec); 1.8 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 40, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Max,qwen3-30b-a3b,Q4_K_M,ctx 4096,71.82 tok/s (estimated, apple-m2-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-max", "tps": 71.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Max", "id": "llmconfigurator-est-apple-m2-max-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 40 W (package-spec); 0.88 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 40, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Max,qwen3-8b,Q4_K_M,ctx 4096,35.29 tok/s (estimated, apple-m2-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-max", "tps": 35.29, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Pro", "id": "llmconfigurator-est-apple-m3-pro-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.5 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Pro,llama-3.1-8b,Q4_K_M,ctx 4096,14.93 tok/s (estimated, apple-m3-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-pro", "tps": 14.93, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Pro", "id": "llmconfigurator-est-apple-m3-pro-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.55 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Pro,qwen-2.5-7b,Q4_K_M,ctx 4096,16.52 tok/s (estimated, apple-m3-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-pro", "tps": 16.52, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Pro", "id": "llmconfigurator-est-apple-m3-pro-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.28 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Pro,qwen3-14b,Q4_K_M,ctx 4096,8.51 tok/s (estimated, apple-m3-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-pro", "tps": 8.51, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Pro", "id": "llmconfigurator-est-apple-m3-pro-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.1 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Pro,qwen3-30b-a3b,Q4_K_M,ctx 4096,31.69 tok/s (estimated, apple-m3-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-pro", "tps": 31.69, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Pro", "id": "llmconfigurator-est-apple-m3-pro-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.48 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Pro,qwen3-8b,Q4_K_M,ctx 4096,14.44 tok/s (estimated, apple-m3-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-pro", "tps": 14.44, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Max", "id": "llmconfigurator-est-apple-m4-max-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.4 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Max,llama-3.1-8b,Q4_K_M,ctx 4096,47.77 tok/s (estimated, apple-m4-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-max", "tps": 47.77, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Max", "id": "llmconfigurator-est-apple-m4-max-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.5 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Max,qwen-2.5-7b,Q4_K_M,ctx 4096,52.31 tok/s (estimated, apple-m4-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-max", "tps": 52.31, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Max", "id": "llmconfigurator-est-apple-m4-max-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.81 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Max,qwen3-14b,Q4_K_M,ctx 4096,28.49 tok/s (estimated, apple-m4-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-max", "tps": 28.49, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Max", "id": "llmconfigurator-est-apple-m4-max-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 2.6 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Max,qwen3-30b-a3b,Q4_K_M,ctx 4096,90.88 tok/s (estimated, apple-m4-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-max", "tps": 90.88, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Max", "id": "llmconfigurator-est-apple-m4-max-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.3 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Max,qwen3-8b,Q4_K_M,ctx 4096,46.38 tok/s (estimated, apple-m4-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-max", "tps": 46.38, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Pro", "id": "llmconfigurator-est-apple-m4-pro-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.86 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Pro,llama-3.1-8b,Q4_K_M,ctx 4096,25.73 tok/s (estimated, apple-m4-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-pro", "tps": 25.73, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Pro", "id": "llmconfigurator-est-apple-m4-pro-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.95 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Pro,qwen-2.5-7b,Q4_K_M,ctx 4096,28.38 tok/s (estimated, apple-m4-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-pro", "tps": 28.38, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Pro", "id": "llmconfigurator-est-apple-m4-pro-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.5 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Pro,qwen3-14b,Q4_K_M,ctx 4096,14.88 tok/s (estimated, apple-m4-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-pro", "tps": 14.88, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4 Pro", "id": "llmconfigurator-est-apple-m4-pro-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.83 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4 Pro,qwen3-8b,Q4_K_M,ctx 4096,24.92 tok/s (estimated, apple-m4-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4-pro", "tps": 24.92, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max", "id": "llmconfigurator-est-apple-m3-max-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max,llama-3.1-8b,Q4_K_M,ctx 4096,36.39 tok/s (estimated, apple-m3-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max", "tps": 36.39, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max", "id": "llmconfigurator-est-apple-m3-max-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 35 W (package-spec); 1.1 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max,qwen-2.5-7b,Q4_K_M,ctx 4096,40 tok/s (estimated, apple-m3-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max", "tps": 40, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max", "id": "llmconfigurator-est-apple-m3-max-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.61 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max,qwen3-14b,Q4_K_M,ctx 4096,21.36 tok/s (estimated, apple-m3-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max", "tps": 21.36, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max", "id": "llmconfigurator-est-apple-m3-max-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 2.1 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max,qwen3-30b-a3b,Q4_K_M,ctx 4096,71.82 tok/s (estimated, apple-m3-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max", "tps": 71.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Max", "id": "llmconfigurator-est-apple-m3-max-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Max,qwen3-8b,Q4_K_M,ctx 4096,35.29 tok/s (estimated, apple-m3-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-max", "tps": 35.29, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1", "id": "llmconfigurator-est-apple-m1-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.34 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1,llama-3.1-8b,Q4_K_M,ctx 4096,6.8 tok/s (estimated, apple-m1)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1", "tps": 6.8, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1", "id": "llmconfigurator-est-apple-m1-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.38 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1,qwen-2.5-7b,Q4_K_M,ctx 4096,7.55 tok/s (estimated, apple-m1)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1", "tps": 7.55, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1", "id": "llmconfigurator-est-apple-m1-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.19 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1,qwen3-14b,Q4_K_M,ctx 4096,3.84 tok/s (estimated, apple-m1)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1", "tps": 3.84, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1", "id": "llmconfigurator-est-apple-m1-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.33 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1,qwen3-8b,Q4_K_M,ctx 4096,6.58 tok/s (estimated, apple-m1)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1", "tps": 6.58, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Pro", "id": "llmconfigurator-est-apple-m1-pro-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.64 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Pro,llama-3.1-8b,Q4_K_M,ctx 4096,19.25 tok/s (estimated, apple-m1-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-pro", "tps": 19.25, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Pro", "id": "llmconfigurator-est-apple-m1-pro-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.71 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Pro,qwen-2.5-7b,Q4_K_M,ctx 4096,21.28 tok/s (estimated, apple-m1-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-pro", "tps": 21.28, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Pro", "id": "llmconfigurator-est-apple-m1-pro-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.37 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Pro,qwen3-14b,Q4_K_M,ctx 4096,11.03 tok/s (estimated, apple-m1-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-pro", "tps": 11.03, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Pro", "id": "llmconfigurator-est-apple-m1-pro-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.3 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Pro,qwen3-30b-a3b,Q4_K_M,ctx 4096,40.24 tok/s (estimated, apple-m1-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-pro", "tps": 40.24, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Pro", "id": "llmconfigurator-est-apple-m1-pro-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.62 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Pro,qwen3-8b,Q4_K_M,ctx 4096,18.63 tok/s (estimated, apple-m1-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-pro", "tps": 18.63, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Ultra", "id": "llmconfigurator-est-apple-m1-ultra-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.1 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Ultra,llama-3.1-8b,Q4_K_M,ctx 4096,65.62 tok/s (estimated, apple-m1-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-ultra", "tps": 65.62, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Ultra", "id": "llmconfigurator-est-apple-m1-ultra-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.2 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Ultra,qwen-2.5-7b,Q4_K_M,ctx 4096,71.43 tok/s (estimated, apple-m1-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-ultra", "tps": 71.43, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Ultra", "id": "llmconfigurator-est-apple-m1-ultra-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 0.67 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Ultra,qwen3-14b,Q4_K_M,ctx 4096,40.15 tok/s (estimated, apple-m1-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-ultra", "tps": 40.15, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Ultra", "id": "llmconfigurator-est-apple-m1-ultra-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 2 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Ultra,qwen3-30b-a3b,Q4_K_M,ctx 4096,118.17 tok/s (estimated, apple-m1-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-ultra", "tps": 118.17, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M1 Ultra", "id": "llmconfigurator-est-apple-m1-ultra-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.1 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M1 Ultra,qwen3-8b,Q4_K_M,ctx 4096,63.82 tok/s (estimated, apple-m1-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m1-ultra", "tps": 63.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2", "id": "llmconfigurator-est-apple-m2-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.5 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2,llama-3.1-8b,Q4_K_M,ctx 4096,9.91 tok/s (estimated, apple-m2)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2", "tps": 9.91, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2", "id": "llmconfigurator-est-apple-m2-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.55 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2,qwen-2.5-7b,Q4_K_M,ctx 4096,10.99 tok/s (estimated, apple-m2)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2", "tps": 10.99, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2", "id": "llmconfigurator-est-apple-m2-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.28 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2,qwen3-14b,Q4_K_M,ctx 4096,5.61 tok/s (estimated, apple-m2)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2", "tps": 5.61, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2", "id": "llmconfigurator-est-apple-m2-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 20 W (package-spec); 0.48 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 20, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2,qwen3-8b,Q4_K_M,ctx 4096,9.58 tok/s (estimated, apple-m2)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2", "tps": 9.58, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Ultra", "id": "llmconfigurator-est-apple-m2-ultra-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.1 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Ultra,llama-3.1-8b,Q4_K_M,ctx 4096,65.62 tok/s (estimated, apple-m2-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-ultra", "tps": 65.62, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Ultra", "id": "llmconfigurator-est-apple-m2-ultra-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.2 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Ultra,qwen-2.5-7b,Q4_K_M,ctx 4096,71.43 tok/s (estimated, apple-m2-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-ultra", "tps": 71.43, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Ultra", "id": "llmconfigurator-est-apple-m2-ultra-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 0.67 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Ultra,qwen3-14b,Q4_K_M,ctx 4096,40.15 tok/s (estimated, apple-m2-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-ultra", "tps": 40.15, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Ultra", "id": "llmconfigurator-est-apple-m2-ultra-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 2 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Ultra,qwen3-30b-a3b,Q4_K_M,ctx 4096,118.17 tok/s (estimated, apple-m2-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-ultra", "tps": 118.17, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M2 Ultra", "id": "llmconfigurator-est-apple-m2-ultra-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.1 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M2 Ultra,qwen3-8b,Q4_K_M,ctx 4096,63.82 tok/s (estimated, apple-m2-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m2-ultra", "tps": 63.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3", "id": "llmconfigurator-est-apple-m3-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.45 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3,llama-3.1-8b,Q4_K_M,ctx 4096,9.91 tok/s (estimated, apple-m3)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3", "tps": 9.91, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3", "id": "llmconfigurator-est-apple-m3-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.5 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3,qwen-2.5-7b,Q4_K_M,ctx 4096,10.99 tok/s (estimated, apple-m3)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3", "tps": 10.99, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3", "id": "llmconfigurator-est-apple-m3-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.25 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3,qwen3-14b,Q4_K_M,ctx 4096,5.61 tok/s (estimated, apple-m3)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3", "tps": 5.61, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3", "id": "llmconfigurator-est-apple-m3-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.44 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3,qwen3-8b,Q4_K_M,ctx 4096,9.58 tok/s (estimated, apple-m3)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3", "tps": 9.58, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Ultra", "id": "llmconfigurator-est-apple-m3-ultra-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.1 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Ultra,llama-3.1-8b,Q4_K_M,ctx 4096,66.86 tok/s (estimated, apple-m3-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-ultra", "tps": 66.86, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Ultra", "id": "llmconfigurator-est-apple-m3-ultra-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.2 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Ultra,qwen-2.5-7b,Q4_K_M,ctx 4096,72.75 tok/s (estimated, apple-m3-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-ultra", "tps": 72.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Ultra", "id": "llmconfigurator-est-apple-m3-ultra-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 0.68 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Ultra,qwen3-14b,Q4_K_M,ctx 4096,40.98 tok/s (estimated, apple-m3-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-ultra", "tps": 40.98, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Ultra", "id": "llmconfigurator-est-apple-m3-ultra-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 2 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Ultra,qwen3-30b-a3b,Q4_K_M,ctx 4096,119.97 tok/s (estimated, apple-m3-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-ultra", "tps": 119.97, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M3 Ultra", "id": "llmconfigurator-est-apple-m3-ultra-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 60 W (package-spec); 1.1 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 60, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M3 Ultra,qwen3-8b,Q4_K_M,ctx 4096,65.04 tok/s (estimated, apple-m3-ultra)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m3-ultra", "tps": 65.04, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4", "id": "llmconfigurator-est-apple-m4-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.54 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4,llama-3.1-8b,Q4_K_M,ctx 4096,11.82 tok/s (estimated, apple-m4)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4", "tps": 11.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4", "id": "llmconfigurator-est-apple-m4-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.6 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4,qwen-2.5-7b,Q4_K_M,ctx 4096,13.1 tok/s (estimated, apple-m4)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4", "tps": 13.1, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4", "id": "llmconfigurator-est-apple-m4-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.3 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4,qwen3-14b,Q4_K_M,ctx 4096,6.71 tok/s (estimated, apple-m4)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4", "tps": 6.71, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4", "id": "llmconfigurator-est-apple-m4-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 1.2 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4,qwen3-30b-a3b,Q4_K_M,ctx 4096,25.37 tok/s (estimated, apple-m4)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4", "tps": 25.37, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M4", "id": "llmconfigurator-est-apple-m4-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 22 W (package-spec); 0.52 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 22, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M4,qwen3-8b,Q4_K_M,ctx 4096,11.43 tok/s (estimated, apple-m4)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m4", "tps": 11.43, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5", "id": "llmconfigurator-est-apple-m5-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 23 W (package-spec); 0.65 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 23, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5,llama-3.1-8b,Q4_K_M,ctx 4096,14.93 tok/s (estimated, apple-m5)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5", "tps": 14.93, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5", "id": "llmconfigurator-est-apple-m5-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 23 W (package-spec); 0.72 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 23, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5,qwen-2.5-7b,Q4_K_M,ctx 4096,16.52 tok/s (estimated, apple-m5)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5", "tps": 16.52, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5", "id": "llmconfigurator-est-apple-m5-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 23 W (package-spec); 0.37 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 23, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5,qwen3-14b,Q4_K_M,ctx 4096,8.51 tok/s (estimated, apple-m5)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5", "tps": 8.51, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5", "id": "llmconfigurator-est-apple-m5-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 23 W (package-spec); 1.4 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 23, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5,qwen3-30b-a3b,Q4_K_M,ctx 4096,31.69 tok/s (estimated, apple-m5)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5", "tps": 31.69, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5", "id": "llmconfigurator-est-apple-m5-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 23 W (package-spec); 0.63 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 23, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5,qwen3-8b,Q4_K_M,ctx 4096,14.44 tok/s (estimated, apple-m5)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5", "tps": 14.44, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Pro", "id": "llmconfigurator-est-apple-m5-pro-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.96 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Pro,llama-3.1-8b,Q4_K_M,ctx 4096,28.66 tok/s (estimated, apple-m5-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-pro", "tps": 28.66, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Pro", "id": "llmconfigurator-est-apple-m5-pro-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.1 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Pro,qwen-2.5-7b,Q4_K_M,ctx 4096,31.58 tok/s (estimated, apple-m5-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-pro", "tps": 31.58, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Pro", "id": "llmconfigurator-est-apple-m5-pro-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.55 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Pro,qwen3-14b,Q4_K_M,ctx 4096,16.64 tok/s (estimated, apple-m5-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-pro", "tps": 16.64, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Pro", "id": "llmconfigurator-est-apple-m5-pro-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 1.9 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Pro,qwen3-30b-a3b,Q4_K_M,ctx 4096,58.03 tok/s (estimated, apple-m5-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-pro", "tps": 58.03, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Pro", "id": "llmconfigurator-est-apple-m5-pro-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 30 W (package-spec); 0.93 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 30, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Pro,qwen3-8b,Q4_K_M,ctx 4096,27.77 tok/s (estimated, apple-m5-pro)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-pro", "tps": 27.77, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Max", "id": "llmconfigurator-est-apple-m5-max-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.5 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Max,llama-3.1-8b,Q4_K_M,ctx 4096,52.78 tok/s (estimated, apple-m5-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-max", "tps": 52.78, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Max", "id": "llmconfigurator-est-apple-m5-max-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.6 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Max,qwen-2.5-7b,Q4_K_M,ctx 4096,57.69 tok/s (estimated, apple-m5-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-max", "tps": 57.69, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Max", "id": "llmconfigurator-est-apple-m5-max-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 0.91 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Max,qwen3-14b,Q4_K_M,ctx 4096,31.7 tok/s (estimated, apple-m5-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-max", "tps": 31.7, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Max", "id": "llmconfigurator-est-apple-m5-max-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 2.8 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Max,qwen3-30b-a3b,Q4_K_M,ctx 4096,98.85 tok/s (estimated, apple-m5-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-max", "tps": 98.85, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "M5 Max", "id": "llmconfigurator-est-apple-m5-max-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 35 W (package-spec); 1.5 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 35, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "M5 Max,qwen3-8b,Q4_K_M,ctx 4096,51.27 tok/s (estimated, apple-m5-max)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/apple-m5-max", "tps": 51.27, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B580", "id": "llmconfigurator-est-arc-b580-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 190 W (board-spec); 0.22 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 190, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B580,llama-3.1-8b,Q4_K_M,ctx 4096,42.34 tok/s (estimated, arc-b580)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b580", "tps": 42.34, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B580", "id": "llmconfigurator-est-arc-b580-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 190 W (board-spec); 0.24 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 190, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B580,qwen-2.5-7b,Q4_K_M,ctx 4096,46.23 tok/s (estimated, arc-b580)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b580", "tps": 46.23, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B580", "id": "llmconfigurator-est-arc-b580-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 190 W (board-spec); 0.13 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 190, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B580,qwen3-14b,Q4_K_M,ctx 4096,25.57 tok/s (estimated, arc-b580)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b580", "tps": 25.57, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B580", "id": "llmconfigurator-est-arc-b580-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 190 W (board-spec); 0.22 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 190, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B580,qwen3-8b,Q4_K_M,ctx 4096,41.15 tok/s (estimated, arc-b580)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b580", "tps": 41.15, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B570", "id": "llmconfigurator-est-arc-b570-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.24 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B570,llama-3.1-8b,Q4_K_M,ctx 4096,36.31 tok/s (estimated, arc-b570)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b570", "tps": 36.31, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B570", "id": "llmconfigurator-est-arc-b570-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.27 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B570,qwen-2.5-7b,Q4_K_M,ctx 4096,39.75 tok/s (estimated, arc-b570)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b570", "tps": 39.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Arc B570", "id": "llmconfigurator-est-arc-b570-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (board-spec); 0.24 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Arc B570,qwen3-8b,Q4_K_M,ctx 4096,35.26 tok/s (estimated, arc-b570)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/arc-b570", "tps": 35.26, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "DGX Spark", "id": "llmconfigurator-est-nvidia-dgx-spark-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (package-spec); 0.26 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DGX Spark,llama-3.1-8b,Q4_K_M,ctx 4096,39.41 tok/s (estimated, nvidia-dgx-spark)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-dgx-spark", "tps": 39.41, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "DGX Spark", "id": "llmconfigurator-est-nvidia-dgx-spark-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (package-spec); 0.29 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DGX Spark,qwen-2.5-7b,Q4_K_M,ctx 4096,43.42 tok/s (estimated, nvidia-dgx-spark)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-dgx-spark", "tps": 43.42, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "DGX Spark", "id": "llmconfigurator-est-nvidia-dgx-spark-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (package-spec); 0.15 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DGX Spark,qwen3-14b,Q4_K_M,ctx 4096,22.91 tok/s (estimated, nvidia-dgx-spark)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-dgx-spark", "tps": 22.91, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "DGX Spark", "id": "llmconfigurator-est-nvidia-dgx-spark-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (package-spec); 0.53 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DGX Spark,qwen3-30b-a3b,Q4_K_M,ctx 4096,79.57 tok/s (estimated, nvidia-dgx-spark)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-dgx-spark", "tps": 79.57, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "DGX Spark", "id": "llmconfigurator-est-nvidia-dgx-spark-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 150 W (package-spec); 0.25 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 150, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "DGX Spark,qwen3-8b,Q4_K_M,ctx 4096,38.19 tok/s (estimated, nvidia-dgx-spark)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-dgx-spark", "tps": 38.19, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI Max+ 395", "id": "llmconfigurator-est-amd-ryzen-ai-max-plus-395-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 120 W (package-spec); 0.23 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI Max+ 395,llama-3.1-8b,Q4_K_M,ctx 4096,27.97 tok/s (estimated, amd-ryzen-ai-max-plus-395)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-max-plus-395", "tps": 27.97, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI Max+ 395", "id": "llmconfigurator-est-amd-ryzen-ai-max-plus-395-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 120 W (package-spec); 0.26 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI Max+ 395,qwen-2.5-7b,Q4_K_M,ctx 4096,30.78 tok/s (estimated, amd-ryzen-ai-max-plus-395)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-max-plus-395", "tps": 30.78, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI Max+ 395", "id": "llmconfigurator-est-amd-ryzen-ai-max-plus-395-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 120 W (package-spec); 0.14 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI Max+ 395,qwen3-14b,Q4_K_M,ctx 4096,16.33 tok/s (estimated, amd-ryzen-ai-max-plus-395)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-max-plus-395", "tps": 16.33, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI Max+ 395", "id": "llmconfigurator-est-amd-ryzen-ai-max-plus-395-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 120 W (package-spec); 0.47 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI Max+ 395,qwen3-30b-a3b,Q4_K_M,ctx 4096,55.89 tok/s (estimated, amd-ryzen-ai-max-plus-395)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-max-plus-395", "tps": 55.89, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "Ryzen AI Max+ 395", "id": "llmconfigurator-est-amd-ryzen-ai-max-plus-395-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 120 W (package-spec); 0.23 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 120, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "Ryzen AI Max+ 395,qwen3-8b,Q4_K_M,ctx 4096,27.11 tok/s (estimated, amd-ryzen-ai-max-plus-395)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/amd-ryzen-ai-max-plus-395", "tps": 27.11, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 6000 Ada Generation", "id": "llmconfigurator-est-rtx-6000-ada-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 300 W (board-spec); 0.4 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 6000 Ada Generation,llama-3.1-8b,Q4_K_M,ctx 4096,120.82 tok/s (estimated, rtx-6000-ada)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-6000-ada", "tps": 120.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 6000 Ada Generation", "id": "llmconfigurator-est-rtx-6000-ada-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 300 W (board-spec); 0.43 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 6000 Ada Generation,qwen-2.5-7b,Q4_K_M,ctx 4096,130.45 tok/s (estimated, rtx-6000-ada)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-6000-ada", "tps": 130.45, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 6000 Ada Generation", "id": "llmconfigurator-est-rtx-6000-ada-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 300 W (board-spec); 0.26 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 6000 Ada Generation,qwen3-14b,Q4_K_M,ctx 4096,76.68 tok/s (estimated, rtx-6000-ada)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-6000-ada", "tps": 76.68, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 6000 Ada Generation", "id": "llmconfigurator-est-rtx-6000-ada-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 300 W (board-spec); 0.68 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 6000 Ada Generation,qwen3-30b-a3b,Q4_K_M,ctx 4096,202.54 tok/s (estimated, rtx-6000-ada)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-6000-ada", "tps": 202.54, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX 6000 Ada Generation", "id": "llmconfigurator-est-rtx-6000-ada-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 300 W (board-spec); 0.39 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX 6000 Ada Generation,qwen3-8b,Q4_K_M,ctx 4096,117.82 tok/s (estimated, rtx-6000-ada)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-6000-ada", "tps": 117.82, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX Pro 6000 Blackwell", "id": "llmconfigurator-est-rtx-pro-6000-blackwell-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 600 W (board-spec); 0.29 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 600, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX PRO 6000 Blackwell,llama-3.1-8b,Q4_K_M,ctx 4096,173.33 tok/s (estimated, rtx-pro-6000-blackwell)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-pro-6000-blackwell", "tps": 173.33, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX Pro 6000 Blackwell", "id": "llmconfigurator-est-rtx-pro-6000-blackwell-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 600 W (board-spec); 0.31 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 600, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX PRO 6000 Blackwell,qwen-2.5-7b,Q4_K_M,ctx 4096,184.75 tok/s (estimated, rtx-pro-6000-blackwell)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-pro-6000-blackwell", "tps": 184.75, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX Pro 6000 Blackwell", "id": "llmconfigurator-est-rtx-pro-6000-blackwell-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 600 W (board-spec); 0.19 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 600, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX PRO 6000 Blackwell,qwen3-14b,Q4_K_M,ctx 4096,116.92 tok/s (estimated, rtx-pro-6000-blackwell)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-pro-6000-blackwell", "tps": 116.92, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX Pro 6000 Blackwell", "id": "llmconfigurator-est-rtx-pro-6000-blackwell-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 600 W (board-spec); 0.44 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 600, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX PRO 6000 Blackwell,qwen3-30b-a3b,Q4_K_M,ctx 4096,261.89 tok/s (estimated, rtx-pro-6000-blackwell)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-pro-6000-blackwell", "tps": 261.89, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "RTX Pro 6000 Blackwell", "id": "llmconfigurator-est-rtx-pro-6000-blackwell-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 600 W (board-spec); 0.28 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 600, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "RTX PRO 6000 Blackwell,qwen3-8b,Q4_K_M,ctx 4096,169.7 tok/s (estimated, rtx-pro-6000-blackwell)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/rtx-pro-6000-blackwell", "tps": 169.7, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "L40S", "id": "llmconfigurator-est-nvidia-l40s-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 350 W (board-spec); 0.32 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "L40S,llama-3.1-8b,Q4_K_M,ctx 4096,111.77 tok/s (estimated, nvidia-l40s)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-l40s", "tps": 111.77, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "L40S", "id": "llmconfigurator-est-nvidia-l40s-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 350 W (board-spec); 0.35 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "L40S,qwen-2.5-7b,Q4_K_M,ctx 4096,120.94 tok/s (estimated, nvidia-l40s)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-l40s", "tps": 120.94, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "L40S", "id": "llmconfigurator-est-nvidia-l40s-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 350 W (board-spec); 0.2 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "L40S,qwen3-14b,Q4_K_M,ctx 4096,70.22 tok/s (estimated, nvidia-l40s)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-l40s", "tps": 70.22, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "L40S", "id": "llmconfigurator-est-nvidia-l40s-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 350 W (board-spec); 0.55 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "L40S,qwen3-30b-a3b,Q4_K_M,ctx 4096,190.97 tok/s (estimated, nvidia-l40s)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-l40s", "tps": 190.97, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "L40S", "id": "llmconfigurator-est-nvidia-l40s-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 12 runs, MAPE 7.68%); power 350 W (board-spec); 0.31 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "L40S,qwen3-8b,Q4_K_M,ctx 4096,108.92 tok/s (estimated, nvidia-l40s)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-l40s", "tps": 108.92, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "A100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-a100-80gb-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 300 W (board-spec); 0.5 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "A100 80GB (PCIe),llama-3.1-8b,Q4_K_M,ctx 4096,149.19 tok/s (estimated, nvidia-a100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-a100-80gb", "tps": 149.19, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "A100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-a100-80gb-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 300 W (board-spec); 0.53 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "A100 80GB (PCIe),qwen-2.5-7b,Q4_K_M,ctx 4096,159.96 tok/s (estimated, nvidia-a100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-a100-80gb", "tps": 159.96, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "A100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-a100-80gb-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 300 W (board-spec); 0.33 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "A100 80GB (PCIe),qwen3-14b,Q4_K_M,ctx 4096,97.81 tok/s (estimated, nvidia-a100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-a100-80gb", "tps": 97.81, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "A100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-a100-80gb-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 300 W (board-spec); 0.79 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "A100 80GB (PCIe),qwen3-30b-a3b,Q4_K_M,ctx 4096,236.13 tok/s (estimated, nvidia-a100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-a100-80gb", "tps": 236.13, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "A100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-a100-80gb-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 1 runs, unfitted); power 300 W (board-spec); 0.49 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 300, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "A100 80GB (PCIe),qwen3-8b,Q4_K_M,ctx 4096,145.79 tok/s (estimated, nvidia-a100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-a100-80gb", "tps": 145.79, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "H100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-h100-80gb-llama-3-1-8b-q4-k-m-ctx4096", "model": "Llama 3.1 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 350 W (board-spec); 0.53 tok/W; VRAM fit great (6.17 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "H100 80GB (PCIe),llama-3.1-8b,Q4_K_M,ctx 4096,185.08 tok/s (estimated, nvidia-h100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-h100-80gb", "tps": 185.08, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "H100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-h100-80gb-qwen-2-5-7b-q4-k-m-ctx4096", "model": "qwen-2.5-7b", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 350 W (board-spec); 0.56 tok/W; VRAM fit great (5.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "H100 80GB (PCIe),qwen-2.5-7b,Q4_K_M,ctx 4096,196.72 tok/s (estimated, nvidia-h100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-h100-80gb", "tps": 196.72, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "H100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-h100-80gb-qwen3-14b-q4-k-m-ctx4096", "model": "Qwen3 14B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 350 W (board-spec); 0.36 tok/W; VRAM fit great (10.41 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "H100 80GB (PCIe),qwen3-14b,Q4_K_M,ctx 4096,126.63 tok/s (estimated, nvidia-h100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-h100-80gb", "tps": 126.63, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "H100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-h100-80gb-qwen3-30b-a3b-q4-k-m-ctx4096", "model": "Qwen3 30B A3B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 350 W (board-spec); 0.78 tok/W; VRAM fit great (19.62 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "H100 80GB (PCIe),qwen3-30b-a3b,Q4_K_M,ctx 4096,273.61 tok/s (estimated, nvidia-h100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-h100-80gb", "tps": 273.61, "ttft_s": null }, { "backend": "llama.cpp", "batch": null, "ctx": 4096, "date": "2026-08-17", "hardware": "H100 80GB (PCIe)", "id": "llmconfigurator-est-nvidia-h100-80gb-qwen3-8b-q4-k-m-ctx4096", "model": "Qwen3 8B", "notes": "bandwidth-model estimate (fitted from 0 runs, unfitted); power 350 W (board-spec); 0.52 tok/W; VRAM fit great (6.35 GB required)", "params": null, "power_w": 350, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K_M", "quote": "H100 80GB (PCIe),qwen3-8b,Q4_K_M,ctx 4096,181.37 tok/s (estimated, nvidia-h100-80gb)", "ram_gb": null, "retrieved": "2026-10-03", "source_name": "LLM Configurator benchmark cells (CC BY 4.0)", "source_url": "https://llmconfigurator.com/en/benchmarks/nvidia-h100-80gb", "tps": 181.37, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-gemma-3-27b-q4-k-medium-ctxnone", "model": "Gemma 3 27B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Gemma 3 27B,Q4_K - Medium,ctx None,30 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-gemma-3-27b-q4-k-medium-ctxnone", "model": "Gemma 3 27B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Gemma 3 27B,Q4_K - Medium,ctx None,14 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 14, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Max (48 GB)", "id": "siliconscore-m4-max-48-gb-gemma-3-27b-q4-k-medium-ctxnone", "model": "Gemma 3 27B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Max (48 GB),Gemma 3 27B,Q4_K - Medium,ctx None,27 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 27, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M3 Max (96 GB)", "id": "siliconscore-m3-max-96-gb-gemma-3-27b-q4-k-medium-ctxnone", "model": "Gemma 3 27B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 Max (96 GB),Gemma 3 27B,Q4_K - Medium,ctx None,20 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-gemma-4-e2b-q4-k-medium-ctxnone", "model": "Gemma 4 E2B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Gemma 4 E2B,Q4_K - Medium,ctx None,261 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 261, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-gemma-4-e2b-q4-k-medium-ctxnone", "model": "Gemma 4 E2B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Gemma 4 E2B,Q4_K - Medium,ctx None,147 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 147, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-gemma-4-e2b-q4-k-medium-ctxnone", "model": "Gemma 4 E2B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Gemma 4 E2B,Q4_K - Medium,ctx None,62 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 62, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (8 GB)", "id": "siliconscore-m1-8-gb-gemma-4-e2b-q4-k-medium-ctxnone", "model": "Gemma 4 E2B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (8 GB),Gemma 4 E2B,Q4_K - Medium,ctx None,44 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 44, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-gemma-4-e4b-q4-k-medium-ctxnone", "model": "Gemma 4 E4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Gemma 4 E4B,Q4_K - Medium,ctx None,161 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Pro (24 GB)", "id": "siliconscore-m5-pro-24-gb-gemma-4-e4b-q4-k-medium-ctxnone", "model": "Gemma 4 E4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Pro (24 GB),Gemma 4 E4B,Q4_K - Medium,ctx None,91 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-gemma-4-e4b-q4-k-medium-ctxnone", "model": "Gemma 4 E4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Gemma 4 E4B,Q4_K - Medium,ctx None,82 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 82, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-gemma-4-e4b-q4-k-medium-ctxnone", "model": "Gemma 4 E4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Gemma 4 E4B,Q4_K - Medium,ctx None,33 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 33, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (8 GB)", "id": "siliconscore-m1-8-gb-gemma-4-e4b-q4-k-medium-ctxnone", "model": "Gemma 4 E4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (8 GB),Gemma 4 E4B,Q4_K - Medium,ctx None,23 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-mistral-small-4-119b-q4-k-medium-ctxnone", "model": "Mistral Small 4 119B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Mistral Small 4 119B,Q4_K - Medium,ctx None,46 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 46, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-llama-3-3-70b-q4-k-medium-ctxnone", "model": "Llama 3.3 70B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Llama 3.3 70B,Q4_K - Medium,ctx None,12 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-deepseek-r1-distill-llama-70b-q4-k-medium-ctxnone", "model": "DeepSeek R1 70B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),DeepSeek R1 70B,Q4_K - Medium,ctx None,12 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-qwen-2-5-72b-q4-k-medium-ctxnone", "model": "Qwen 2.5 72B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Qwen 2.5 72B,Q4_K - Medium,ctx None,12 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-qwen-3-235b-a22b-q4-k-medium-ctxnone", "model": "Qwen 3 235B-A22B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Qwen 3 235B-A22B,Q4_K - Medium,ctx None,15 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-nemotron-cascade-2-30b-a3b-q4-k-medium-ctxnone", "model": "Nemotron Cascade 2 30B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Nemotron Cascade 2 30B-A3B,Q4_K - Medium,ctx None,117 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 117, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Max (48 GB)", "id": "siliconscore-m4-max-48-gb-nemotron-cascade-2-30b-a3b-q4-k-medium-ctxnone", "model": "Nemotron Cascade 2 30B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Max (48 GB),Nemotron Cascade 2 30B-A3B,Q4_K - Medium,ctx None,111 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 111, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-nemotron-cascade-2-30b-a3b-q4-k-medium-ctxnone", "model": "Nemotron Cascade 2 30B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Nemotron Cascade 2 30B-A3B,Q4_K - Medium,ctx None,77 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 77, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-llama-4-scout-17b-16e-q4-k-medium-ctxnone", "model": "Llama 4 Scout 17B-16E", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Llama 4 Scout 17B-16E,Q4_K - Medium,ctx None,19 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 19, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-gpt-oss-120b-q4-k-medium-ctxnone", "model": "gpt-oss 120B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),gpt-oss 120B,Q4_K - Medium,ctx None,58 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 58, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Pro (24 GB)", "id": "siliconscore-m5-pro-24-gb-gemma-4-26b-a4b-q4-k-medium-ctxnone", "model": "Gemma 4 26B-A4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Pro (24 GB),Gemma 4 26B-A4B,Q4_K - Medium,ctx None,72 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 72, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-gemma-3-4b-q4-k-medium-ctxnone", "model": "Gemma 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Gemma 3 4B,Q4_K - Medium,ctx None,161 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (8 GB)", "id": "siliconscore-m1-8-gb-gemma-3-4b-q4-k-medium-ctxnone", "model": "Gemma 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (8 GB),Gemma 3 4B,Q4_K - Medium,ctx None,23 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M3 Pro (18 GB)", "id": "siliconscore-m3-pro-18-gb-gemma-3-4b-q4-k-medium-ctxnone", "model": "Gemma 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 Pro (18 GB),Gemma 3 4B,Q4_K - Medium,ctx None,48 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 48, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-qwen3-5-9b-q4-k-medium-ctxnone", "model": "Qwen3.5-9B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Qwen3.5-9B,Q4_K - Medium,ctx None,82 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 82, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": null, "hardware": "M4 (16 GB)", "id": "siliconscore-m4-16-gb-qwen3-5-9b-q4-k-medium-ctxnone", "model": "Qwen3.5-9B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 (16 GB),Qwen3.5-9B,Q4_K - Medium,ctx None,18 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 18, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-qwen3-5-9b-q4-k-medium-ctxnone", "model": "Qwen3.5-9B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Qwen3.5-9B,Q4_K - Medium,ctx None,15 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 15, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (16 GB)", "id": "siliconscore-m1-16-gb-qwen3-5-9b-q4-k-medium-ctxnone", "model": "Qwen3.5-9B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (16 GB),Qwen3.5-9B,Q4_K - Medium,ctx None,10 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-qwen-3-8b-q4-k-medium-ctxnone", "model": "Qwen 3 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Qwen 3 8B,Q4_K - Medium,ctx None,91 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M2 (16 GB)", "id": "siliconscore-m2-16-gb-qwen-3-8b-q4-k-medium-ctxnone", "model": "Qwen 3 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (16 GB),Qwen 3 8B,Q4_K - Medium,ctx None,17 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-phi-4-14b-q4-k-medium-ctxnone", "model": "Phi-4 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Phi-4 14B,Q4_K - Medium,ctx None,55 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 55, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 (16 GB)", "id": "siliconscore-m4-16-gb-phi-4-14b-q4-k-medium-ctxnone", "model": "Phi-4 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 (16 GB),Phi-4 14B,Q4_K - Medium,ctx None,12 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M2 (16 GB)", "id": "siliconscore-m2-16-gb-phi-4-14b-q4-k-medium-ctxnone", "model": "Phi-4 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (16 GB),Phi-4 14B,Q4_K - Medium,ctx None,10 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-qwen-3-30b-a3b-q4-k-medium-ctxnone", "model": "Qwen 3 30B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Qwen 3 30B-A3B,Q4_K - Medium,ctx None,116 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 116, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-qwen-3-30b-a3b-q4-k-medium-ctxnone", "model": "Qwen 3 30B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Qwen 3 30B-A3B,Q4_K - Medium,ctx None,76 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 76, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 Max (36 GB)", "id": "siliconscore-m3-max-36-gb-qwen-3-30b-a3b-q4-k-medium-ctxnone", "model": "Qwen 3 30B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 Max (36 GB),Qwen 3 30B-A3B,Q4_K - Medium,ctx None,94 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 94, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-qwen3-5-35b-a3b-q4-k-medium-ctxnone", "model": "Qwen3.5-35B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Qwen3.5-35B-A3B,Q4_K - Medium,ctx None,117 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 117, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Max (48 GB)", "id": "siliconscore-m4-max-48-gb-qwen3-5-35b-a3b-q4-k-medium-ctxnone", "model": "Qwen3.5-35B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Max (48 GB),Qwen3.5-35B-A3B,Q4_K - Medium,ctx None,111 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 111, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-qwen-3-32b-q4-k-medium-ctxnone", "model": "Qwen 3 32B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Qwen 3 32B,Q4_K - Medium,ctx None,25 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Max (64 GB)", "id": "siliconscore-m4-max-64-gb-qwen-3-32b-q4-k-medium-ctxnone", "model": "Qwen 3 32B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Max (64 GB),Qwen 3 32B,Q4_K - Medium,ctx None,23 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (32 GB)", "id": "siliconscore-m4-pro-32-gb-qwen-3-32b-q4-k-medium-ctxnone", "model": "Qwen 3 32B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (32 GB),Qwen 3 32B,Q4_K - Medium,ctx None,12 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 12, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-deepseek-r1-distill-qwen-32b-q4-k-medium-ctxnone", "model": "DeepSeek R1 32B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),DeepSeek R1 32B,Q4_K - Medium,ctx None,25 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": null, "hardware": "M4 Max (48 GB)", "id": "siliconscore-m4-max-48-gb-deepseek-r1-distill-qwen-32b-q4-k-medium-ctxnone", "model": "DeepSeek R1 32B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Max (48 GB),DeepSeek R1 32B,Q4_K - Medium,ctx None,23 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 23, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 Max (36 GB)", "id": "siliconscore-m3-max-36-gb-deepseek-r1-distill-qwen-32b-q4-k-medium-ctxnone", "model": "DeepSeek R1 32B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 Max (36 GB),DeepSeek R1 32B,Q4_K - Medium,ctx None,17 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 Max (96 GB)", "id": "siliconscore-m3-max-96-gb-qwen3-5-35b-a3b-q4-k-medium-ctxnone", "model": "Qwen3.5-35B-A3B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 Max (96 GB),Qwen3.5-35B-A3B,Q4_K - Medium,ctx None,96 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 96, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-qwen-3-8b-q4-k-medium-ctxnone", "model": "Qwen 3 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Qwen 3 8B,Q4_K - Medium,ctx None,44 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 44, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (16 GB)", "id": "siliconscore-m1-16-gb-qwen-3-8b-q4-k-medium-ctxnone", "model": "Qwen 3 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (16 GB),Qwen 3 8B,Q4_K - Medium,ctx None,11 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-qwen3-5-9b-q4-k-medium-ctxnone", "model": "Qwen3.5-9B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Qwen3.5-9B,Q4_K - Medium,ctx None,39 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 39, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M2 (8 GB)", "id": "siliconscore-m2-8-gb-qwen-3-4b-q4-k-medium-ctxnone", "model": "Qwen 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (8 GB),Qwen 3 4B,Q4_K - Medium,ctx None,33 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 33, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-qwen-3-4b-q4-k-medium-ctxnone", "model": "Qwen 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Qwen 3 4B,Q4_K - Medium,ctx None,82 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 82, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-mistral-7b-v0-3-q4-k-medium-ctxnone", "model": "Mistral 7B v0.3", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Mistral 7B v0.3,Q4_K - Medium,ctx None,102 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 102, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-mistral-7b-v0-3-q4-k-medium-ctxnone", "model": "Mistral 7B v0.3", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Mistral 7B v0.3,Q4_K - Medium,ctx None,50 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 50, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-mistral-7b-v0-3-q4-k-medium-ctxnone", "model": "Mistral 7B v0.3", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Mistral 7B v0.3,Q4_K - Medium,ctx None,19 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 19, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (16 GB)", "id": "siliconscore-m1-16-gb-mistral-7b-v0-3-q4-k-medium-ctxnone", "model": "Mistral 7B v0.3", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (16 GB),Mistral 7B v0.3,Q4_K - Medium,ctx None,13 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 13, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-llama-3-1-8b-q4-k-medium-ctxnone", "model": "Llama 3.1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (128 GB),Llama 3.1 8B,Q4_K - Medium,ctx None,91 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 (16 GB)", "id": "siliconscore-m4-16-gb-llama-3-1-8b-q4-k-medium-ctxnone", "model": "Llama 3.1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 (16 GB),Llama 3.1 8B,Q4_K - Medium,ctx None,20 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 Pro (18 GB)", "id": "siliconscore-m3-pro-18-gb-llama-3-1-8b-q4-k-medium-ctxnone", "model": "Llama 3.1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 Pro (18 GB),Llama 3.1 8B,Q4_K - Medium,ctx None,25 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 25, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (16 GB)", "id": "siliconscore-m1-16-gb-llama-3-1-8b-q4-k-medium-ctxnone", "model": "Llama 3.1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (16 GB),Llama 3.1 8B,Q4_K - Medium,ctx None,11 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M2 (8 GB)", "id": "siliconscore-m2-8-gb-llama-3-1-8b-q4-k-medium-ctxnone", "model": "Llama 3.1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (8 GB),Llama 3.1 8B,Q4_K - Medium,ctx None,17 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-qwen-3-4b-q4-k-medium-ctxnone", "model": "Qwen 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Qwen 3 4B,Q4_K - Medium,ctx None,161 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-qwen3-5-4b-q4-k-medium-ctxnone", "model": "Qwen3.5-4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Qwen3.5-4B,Q4_K - Medium,ctx None,161 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 161, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 (16 GB)", "id": "siliconscore-m4-16-gb-qwen3-5-4b-q4-k-medium-ctxnone", "model": "Qwen3.5-4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 (16 GB),Qwen3.5-4B,Q4_K - Medium,ctx None,39 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 39, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-qwen-3-8b-q8-0-ctxnone", "model": "Qwen 3 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M4 Pro (24 GB),Qwen 3 8B,Q8_0,ctx None,24 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 24, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-mistral-7b-v0-3-q8-0-ctxnone", "model": "Mistral 7B v0.3", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M5 Max (64 GB),Mistral 7B v0.3,Q8_0,ctx None,59 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 59, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-mistral-7b-v0-3-q8-0-ctxnone", "model": "Mistral 7B v0.3", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M4 Pro (24 GB),Mistral 7B v0.3,Q8_0,ctx None,28 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 28, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-qwen3-5-9b-q8-0-ctxnone", "model": "Qwen3.5-9B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M5 Max (128 GB),Qwen3.5-9B,Q8_0,ctx None,47 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 47, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-llama-3-1-8b-q8-0-ctxnone", "model": "Llama 3.1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M5 Max (64 GB),Llama 3.1 8B,Q8_0,ctx None,52 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 52, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-gemma-3-4b-q8-0-ctxnone", "model": "Gemma 3 4B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M4 Pro (24 GB),Gemma 3 4B,Q8_0,ctx None,47 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 47, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-qwen-3-14b-q4-k-medium-ctxnone", "model": "Qwen 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Qwen 3 14B,Q4_K - Medium,ctx None,55 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 55, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-qwen-3-14b-q4-k-medium-ctxnone", "model": "Qwen 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Qwen 3 14B,Q4_K - Medium,ctx None,26 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-qwen-3-14b-q4-k-medium-ctxnone", "model": "Qwen 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Qwen 3 14B,Q4_K - Medium,ctx None,10 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (128 GB)", "id": "siliconscore-m5-max-128-gb-qwen-3-14b-q8-0-ctxnone", "model": "Qwen 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M5 Max (128 GB),Qwen 3 14B,Q8_0,ctx None,31 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 31, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-phi-4-mini-instruct-3-8b-q4-k-medium-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Phi-4 Mini Instruct 3.8B,Q4_K - Medium,ctx None,168 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 168, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-phi-4-mini-instruct-3-8b-q4-k-medium-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Phi-4 Mini Instruct 3.8B,Q4_K - Medium,ctx None,86 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 86, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-phi-4-mini-instruct-3-8b-q4-k-medium-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Phi-4 Mini Instruct 3.8B,Q4_K - Medium,ctx None,34 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 34, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M2 (8 GB)", "id": "siliconscore-m2-8-gb-phi-4-mini-instruct-3-8b-q4-k-medium-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (8 GB),Phi-4 Mini Instruct 3.8B,Q4_K - Medium,ctx None,34 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 34, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (16 GB)", "id": "siliconscore-m1-16-gb-phi-4-mini-instruct-3-8b-q4-k-medium-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (16 GB),Phi-4 Mini Instruct 3.8B,Q4_K - Medium,ctx None,24 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 24, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-phi-4-mini-instruct-3-8b-q8-0-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M5 Max (64 GB),Phi-4 Mini Instruct 3.8B,Q8_0,ctx None,102 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 102, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Max (48 GB)", "id": "siliconscore-m4-max-48-gb-phi-4-mini-instruct-3-8b-q4-k-medium-ctxnone", "model": "Phi-4 Mini Instruct 3.8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Max (48 GB),Phi-4 Mini Instruct 3.8B,Q4_K - Medium,ctx None,153 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 153, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-gemma-3-12b-q4-k-medium-ctxnone", "model": "Gemma 3 12B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Gemma 3 12B,Q4_K - Medium,ctx None,64 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 64, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-gemma-3-12b-q4-k-medium-ctxnone", "model": "Gemma 3 12B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Gemma 3 12B,Q4_K - Medium,ctx None,30 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 30, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-gemma-3-12b-q4-k-medium-ctxnone", "model": "Gemma 3 12B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Gemma 3 12B,Q4_K - Medium,ctx None,11 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M2 (16 GB)", "id": "siliconscore-m2-16-gb-gemma-3-12b-q4-k-medium-ctxnone", "model": "Gemma 3 12B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (16 GB),Gemma 3 12B,Q4_K - Medium,ctx None,11 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-deepseek-r1-distill-llama-8b-q4-k-medium-ctxnone", "model": "DeepSeek R1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),DeepSeek R1 8B,Q4_K - Medium,ctx None,91 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 91, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M4 (16 GB)", "id": "siliconscore-m4-16-gb-deepseek-r1-distill-llama-8b-q4-k-medium-ctxnone", "model": "DeepSeek R1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 (16 GB),DeepSeek R1 8B,Q4_K - Medium,ctx None,20 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 20, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M2 (16 GB)", "id": "siliconscore-m2-16-gb-deepseek-r1-distill-llama-8b-q4-k-medium-ctxnone", "model": "DeepSeek R1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M2 (16 GB),DeepSeek R1 8B,Q4_K - Medium,ctx None,17 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 17, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M1 (16 GB)", "id": "siliconscore-m1-16-gb-deepseek-r1-distill-llama-8b-q4-k-medium-ctxnone", "model": "DeepSeek R1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M1 (16 GB),DeepSeek R1 8B,Q4_K - Medium,ctx None,11 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 11, "ttft_s": null }, { "backend": "MLX", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-deepseek-r1-distill-llama-8b-q8-0-ctxnone", "model": "DeepSeek R1 8B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q8_0", "quote": "M5 Max (64 GB),DeepSeek R1 8B,Q8_0,ctx None,52 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 52, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M5 Max (64 GB)", "id": "siliconscore-m5-max-64-gb-ministral-3-14b-q4-k-medium-ctxnone", "model": "Ministral 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M5 Max (64 GB),Ministral 3 14B,Q4_K - Medium,ctx None,55 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 55, "ttft_s": null }, { "backend": "Ollama", "batch": null, "ctx": null, "date": null, "hardware": "M4 Pro (24 GB)", "id": "siliconscore-m4-pro-24-gb-ministral-3-14b-q4-k-medium-ctxnone", "model": "Ministral 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M4 Pro (24 GB),Ministral 3 14B,Q4_K - Medium,ctx None,26 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 26, "ttft_s": null }, { "backend": "LM Studio", "batch": null, "ctx": null, "date": null, "hardware": "M3 (16 GB)", "id": "siliconscore-m3-16-gb-ministral-3-14b-q4-k-medium-ctxnone", "model": "Ministral 3 14B", "notes": "source label: reference run; verified=False", "params": null, "power_w": null, "pp_tps": null, "provenance": "estimated", "quant": "Q4_K - Medium", "quote": "M3 (16 GB),Ministral 3 14B,Q4_K - Medium,ctx None,10 tok/s", "ram_gb": null, "retrieved": "2026-10-06", "source_name": "Silicon Score benchmark audit", "source_url": "https://llmcheck.net/data/benchmarks.json", "tps": 10, "ttft_s": null } ], "retrieved": "2026-10-22" }