M2 Ultra · 128 GB
800 GB/s memory bandwidth · 108.8 GB usable for models · 19.2 GB left to macOS
The short answer
37 models run comfortably and 0 more will run tight at 8K context. The largest that fits is GPT-OSS 120B at q4, generating around 229 tokens per second.
macOS19.2 GBweights65.7 GBKV cache0.6 GBheadroom42.5 GB
Every model on this Mac
| Model | Size | Best quant | Verdict | Speed est. |
|---|---|---|---|---|
| Qwen2.5 0.5B Instruct | 490M | bf16 | FITS · 107.7 GB FREE | 669 tok/s |
| BGE-M3 | 567M | bf16 | FITS · 107.7 GB FREE | 578 tok/s |
| Qwen3 0.6B | 600M | bf16 | FITS · 106.7 GB FREE | 547 tok/s |
| Parakeet TDT 0.6B v2 | 600M | bf16 | FITS · 107.6 GB FREE | 547 tok/s |
| Llama 3.2 1B Instruct | 1.2B | bf16 | FITS · 106.1 GB FREE | 265 tok/s |
| Whisper Large v3 | 1.6B | bf16 | FITS · 105.7 GB FREE | 212 tok/s |
| Qwen3 1.7B | 1.7B | bf16 | FITS · 104.5 GB FREE | 193 tok/s |
| Llama 3.2 3B Instruct | 3.2B | bf16 | FITS · 101.9 GB FREE | 102 tok/s |
| Qwen3 4B | 4B | q8 | FITS · 103.3 GB FREE | 154 tok/s |
| Qwen3 Embedding 4B | 4B | q8 | FITS · 103.3 GB FREE | 154 tok/s |
| Mistral 7B Instruct v0.3 | 7.3B | q8 | FITS · 100.0 GB FREE | 84.6 tok/s |
| Qwen2.5 7B Instruct | 7.6B | bf16 | FITS · 93.1 GB FREE | 43.2 tok/s |
| InternVL3 8B | 8B | q8 | FITS · 100.3 GB FREE | 77.2 tok/s |
| Llama 3.1 8B Instruct | 8.0B | bf16 | FITS · 91.7 GB FREE | 40.8 tok/s |
| Qwen3 8B | 8.2B | q8 | FITS · 98.9 GB FREE | 75.3 tok/s |
| Qwen2.5-VL 7B Instruct | 8.3B | q8 | FITS · 99.5 GB FREE | 74.4 tok/s |
| Gemma 2 9B IT | 9.2B | q8 | FITS · 99.0 GB FREE | 66.8 tok/s |
| Llama 3.2 11B Vision Instruct | 10.7B | bf16 | FITS · 87.4 GB FREE | 30.7 tok/s |
| Qwen2.5 14B Instruct | 14.7B | q8 | FITS · 91.6 GB FREE | 42.0 tok/s |
| Qwen3 14B | 14.8B | q8 | FITS · 91.7 GB FREE | 41.7 tok/s |
| GPT-OSS 20B | 21B | q8 | FITS · 86.1 GB FREE | 172 tok/s |
| Codestral 22B v0.1 | 22.2B | q8 | FITS · 83.3 GB FREE | 27.8 tok/s |
| Mistral Small 24B Instruct 2501 | 24B | q8 | FITS · 82.0 GB FREE | 25.7 tok/s |
| Gemma 3 27B IT | 27B | bf16 | FITS · 54.8 GB FREE | 12.1 tok/s |
| Gemma 2 27B IT | 27.2B | q8 | FITS · 79.9 GB FREE | 22.7 tok/s |
| Qwen3 30B-A3B Instruct 2507 | 30.5B | q8 | FITS · 75.6 GB FREE | 187 tok/s |
| Qwen3 Coder 30B-A3B Instruct | 30.5B | q8 | FITS · 75.6 GB FREE | 187 tok/s |
| Qwen2.5 32B Instruct | 32.5B | q8 | FITS · 72.1 GB FREE | 19.0 tok/s |
| Qwen2.5 Coder 32B Instruct | 32.5B | q8 | FITS · 72.1 GB FREE | 19.0 tok/s |
| DeepSeek R1 Distill Qwen 32B | 32.5B | q8 | FITS · 72.1 GB FREE | 19.0 tok/s |
| Qwen3 32B | 32.8B | q8 | FITS · 71.8 GB FREE | 18.8 tok/s |
| Qwen2.5-VL 32B Instruct | 33B | q8 | FITS · 71.6 GB FREE | 18.7 tok/s |
| Mixtral 8x7B Instruct v0.1 | 46.7B | q8 | FITS · 58.1 GB FREE | 47.9 tok/s |
| Llama 3.3 70B Instruct | 70.6B | q8 | FITS · 31.1 GB FREE | 8.7 tok/s |
| Qwen2.5 72B Instruct | 72.7B | q8 | FITS · 28.9 GB FREE | 8.5 tok/s |
| Llama 3.2 90B Vision Instruct | 88B | q4 | FITS · 59.3 GB FREE | 13.3 tok/s |
| GPT-OSS 120B | 116.8B | q4 | FITS · 42.5 GB FREE | 229 tok/s |
| Qwen3 235B-A22B | 235B | — | OVER | — |
| DeepSeek V3 | 671B | — | OVER | — |