M3 Ultra (512 GB)
16 records across 7 models (quants: 4bit, 8bit, bf16, q4.1bit).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| M3 Ultra (512 GB) | altar-1 | 4bit | MLX | 13.7 | 86 | – | – | 128 | 2026-10-07 | community | – | source |
| M3 Ultra (512 GB) | altar-1 | 4bit | MLX | 13.0 | 194 | – | – | 4096 | 2026-10-07 | community | – | source |
| M3 Ultra (512 GB) | altar-1 | 4bit | MLX | 12.6 | 205 | – | – | 2048 | 2026-10-07 | community | – | source |
| M3 Ultra (512 GB) | Devstral Small 1.1 | 4bit | LM Studio | 43 | – | – | – | – | – | sourced | – | source |
| M3 Ultra (512 GB) | Gemma 3 27B | bf16 | LM Studio | 11.19 | – | – | 1.72 | 128000 | – | sourced | – | source |
| M3 Ultra (512 GB) | GLM-5 | 4bit | MLX | 16.7 | 187 | – | 5.477 | 1024 | – | sourced | – | source |
| M3 Ultra (512 GB) | GLM-5 | 4bit | MLX | 13.7 | 180.1 | – | 22.745 | 4096 | – | sourced | – | source |
| M3 Ultra (512 GB) | GLM-5 | 4bit | MLX | 13.2 | 154.1 | – | 53.169 | 8192 | – | sourced | – | source |
| M3 Ultra (512 GB) | GLM-5 | 4bit | MLX | 12 | 117.4 | – | 139.545 | 16384 | – | sourced | – | source |
| M3 Ultra (512 GB) | GLM-5 | 4bit | MLX | 10.7 | 77.7 | – | 421.955 | 32768 | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 4bit | LM Studio | 15.5 | 150 | – | – | 7800 | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 4bit | LM Studio | 9.57 | 103 | – | – | 39686 | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 8bit | LM Studio | 8.5 | 150 | – | – | 7800 | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 8bit | LM Studio | 6.53 | 101 | – | – | 39686 | – | sourced | – | source |
| M3 Ultra (512 GB) | Qwen 3 235B-A22B | 4bit | LM Studio | 27.36 | – | – | 1.73 | – | – | sourced | – | source |
| M3 Ultra (512 GB) | Qwen3.5-397B-A17B | q4.1bit | MLX | 40.2 | – | – | – | – | – | sourced | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085