Qwen 3 30B-A3B
13 records across 5 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| M4 Max | Qwen 3 30B-A3B 30B-A3B | 4-bit | MLX | 110 | – | – | – | – | 2026-01 | sourced | – | source |
| M4 Max | Qwen 3 30B-A3B | 4bit | oMLX | 108.4 | 1276.0 | – | – | 1024 | 2026-10-11 | community | – | source |
| M4 Max | Qwen 3 30B-A3B | 4bit | oMLX | 108.4 | 1276.0 | – | – | 1024 | 2026-10-11 | community | – | source |
| M4 Max | Qwen 3 30B-A3B | 4bit | oMLX | 94.5 | 1331.0 | – | – | 4096 | 2026-10-11 | community | – | source |
| M4 Max | Qwen 3 30B-A3B | 4bit | oMLX | 94.5 | 1331.0 | – | – | 4096 | 2026-10-11 | community | – | source |
| M4 Max (40-core GPU, 64 GB) | Qwen 3 30B-A3B | Q4 | MLX | 92.09 | 822.59 | – | – | 2048 | – | sourced | – | source |
| M4 Max | Qwen 3 30B-A3B 30B-A3B | 4-bit | llama.cpp | 90 | – | – | – | – | 2026-01 | sourced | – | source |
| M4 Max (40-core GPU, 64 GB) | Qwen 3 30B-A3B | Q5 | MLX | 84.89 | 819.83 | – | – | 2048 | – | sourced | – | source |
| M4 Max (40-core GPU, 64 GB) | Qwen 3 30B-A3B | Q6 | MLX | 76.74 | 817.63 | – | – | 2048 | – | sourced | – | source |
| M4 Max (128 GB) | Qwen 3 30B-A3B | Q4_K - Medium | LM Studio | 70.15 | – | – | 0.62 | 10000 | – | sourced | – | source |
| M4 Pro (48 GB) | Qwen 3 30B-A3B | 8bit | MLX | 55 | – | – | – | – | – | sourced | – | source |
| M4 Max (40-core GPU, 64 GB) | Qwen 3 30B-A3B | Q8 | MLX | 52.58 | 772.59 | – | – | 2048 | – | sourced | – | source |
| M4 Max (48 GB) | Qwen 3 30B-A3B | Q4_K - Medium | Ollama | 42 | – | – | 0.9 | – | – | sourced | – | source |