M4 · 32 GB
120 GB/s memory bandwidth · 23.0 GB usable for models · 9.0 GB left to macOS
The short answer
27 models run comfortably and 5 more will run tight at 8K context. The largest that fits is Qwen3 30B-A3B Instruct 2507 at q4, generating around 47.8 tokens per second.
macOS9.0 GBweights17.2 GBKV cache0.8 GBheadroom5.1 GB
Every model on this Mac
| Model | Size | Best quant | Verdict | Speed est. |
|---|---|---|---|---|
| Qwen2.5 0.5B Instruct | 490M | bf16 | FITS · 22.0 GB FREE | 90.6 tok/s |
| BGE-M3 | 567M | bf16 | FITS · 21.9 GB FREE | 78.3 tok/s |
| Qwen3 0.6B | 600M | bf16 | FITS · 20.9 GB FREE | 74.0 tok/s |
| Parakeet TDT 0.6B v2 | 600M | bf16 | FITS · 21.8 GB FREE | 74.0 tok/s |
| Llama 3.2 1B Instruct | 1.2B | bf16 | FITS · 20.3 GB FREE | 35.8 tok/s |
| Whisper Large v3 | 1.6B | bf16 | FITS · 19.9 GB FREE | 28.6 tok/s |
| Qwen3 1.7B | 1.7B | bf16 | FITS · 18.7 GB FREE | 26.1 tok/s |
| Llama 3.2 3B Instruct | 3.2B | bf16 | FITS · 16.2 GB FREE | 13.8 tok/s |
| Qwen3 4B | 4B | q8 | FITS · 17.6 GB FREE | 20.9 tok/s |
| Qwen3 Embedding 4B | 4B | q8 | FITS · 17.6 GB FREE | 20.9 tok/s |
| Mistral 7B Instruct v0.3 | 7.3B | q8 | FITS · 14.2 GB FREE | 11.4 tok/s |
| Qwen2.5 7B Instruct | 7.6B | bf16 | FITS · 7.4 GB FREE | 5.8 tok/s |
| InternVL3 8B | 8B | q8 | FITS · 14.5 GB FREE | 10.4 tok/s |
| Llama 3.1 8B Instruct | 8.0B | bf16 | FITS · 5.9 GB FREE | 5.5 tok/s |
| Qwen3 8B | 8.2B | q8 | FITS · 13.1 GB FREE | 10.2 tok/s |
| Qwen2.5-VL 7B Instruct | 8.3B | q8 | FITS · 13.8 GB FREE | 10.1 tok/s |
| Gemma 2 9B IT | 9.2B | q8 | FITS · 13.2 GB FREE | 9.0 tok/s |
| Llama 3.2 11B Vision Instruct | 10.7B | q4 | FITS · 17.0 GB FREE | 14.8 tok/s |
| Qwen2.5 14B Instruct | 14.7B | q8 | FITS · 5.8 GB FREE | 5.7 tok/s |
| Qwen3 14B | 14.8B | q8 | FITS · 6.0 GB FREE | 5.6 tok/s |
| GPT-OSS 20B | 21B | q4 | FITS · 10.8 GB FREE | 43.9 tok/s |
| Codestral 22B v0.1 | 22.2B | q4 | FITS · 8.7 GB FREE | 7.1 tok/s |
| Mistral Small 24B Instruct 2501 | 24B | q4 | FITS · 8.2 GB FREE | 6.6 tok/s |
| Gemma 3 27B IT | 27B | q4 | FITS · 7.9 GB FREE | 5.8 tok/s |
| Gemma 2 27B IT | 27.2B | q4 | FITS · 7.7 GB FREE | 5.8 tok/s |
| Qwen3 30B-A3B Instruct 2507 | 30.5B | q4 | FITS · 5.1 GB FREE | 47.8 tok/s |
| Qwen3 Coder 30B-A3B Instruct | 30.5B | q4 | FITS · 5.1 GB FREE | 47.8 tok/s |
| Qwen2.5 32B Instruct | 32.5B | q4 | TIGHT · 2.6 GB FREE | 4.9 tok/s |
| Qwen2.5 Coder 32B Instruct | 32.5B | q4 | TIGHT · 2.6 GB FREE | 4.9 tok/s |
| DeepSeek R1 Distill Qwen 32B | 32.5B | q4 | TIGHT · 2.6 GB FREE | 4.9 tok/s |
| Qwen3 32B | 32.8B | q4 | TIGHT · 2.4 GB FREE | 4.8 tok/s |
| Qwen2.5-VL 32B Instruct | 33B | q4 | TIGHT · 2.3 GB FREE | 4.8 tok/s |
| Mixtral 8x7B Instruct v0.1 | 46.7B | — | OVER | — |
| Llama 3.3 70B Instruct | 70.6B | — | OVER | — |
| Qwen2.5 72B Instruct | 72.7B | — | OVER | — |
| Llama 3.2 90B Vision Instruct | 88B | — | OVER | — |
| GPT-OSS 120B | 116.8B | — | OVER | — |
| Qwen3 235B-A22B | 235B | — | OVER | — |
| DeepSeek V3 | 671B | — | OVER | — |