| quant | needs | speed |
|---|---|---|
| ~ Q8_0 | 14.0 GB | ~0.5–0.9 tok/s (slow) |
| ~ Q6_K | 11.1 GB | ~0.7–1 tok/s (slow) |
| ~ Q5_K_M | 9.9 GB | ~0.7–1 tok/s (slow) |
| ~ Q4_K_M | 8.7 GB | ~0.9–1 tok/s (slow) |
Same model, other machines: Apple M1 · Apple M1 Pro · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro · Apple M3 Max
Also runs on a Apple M2: DeepSeek-R1 (distill) · Qwen3 · Llama 3.1 · Qwen2.5 · Gemma 3
Best models for a Apple M2 · 16GB →
These share Mistral Nemo's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M2 · 16GB too:
| model | shares | verdict here |
|---|---|---|
| Qwen2.5-Coder | 8kv/128hd | ✅ Runs (tight) |
| Llama 3.2 | 8kv/128hd | ✅ Runs comfortably |
| Llama 3.3 | 8kv/128hd | ❌ Won't fit |
| Mistral 7B | 8kv/128hd | ✅ Runs (tight) |
| Mistral Small 3 | 8kv/128hd | 🐢 CPU-only (slow) |
| Codestral | 8kv/128hd | 🐢 CPU-only (slow) |
| Llama 3.2 Vision | 8kv/128hd | ✅ Runs (tight) |
| Mixtral 8x7B | 8kv/128hd | ❌ Won't fit |
…and 7 more — see the Mistral Nemo page.
Check any combo yourself: open the checker →