| quant | needs | speed |
|---|---|---|
| ~ Q8_0 | 15.8 GB | ~0.6–1 tok/s (slow) |
| ~ Q6_K | 13.0 GB | ~0.8–1 tok/s (slow) |
| ~ Q5_K_M | 11.7 GB | ~0.9–1 tok/s (slow) |
| ~ Q4_K_M | 10.6 GB | ~1–2 tok/s (slow) |
Same model, other machines: Apple M1 · Apple M1 Pro · Apple M2 · Apple M2 Pro · Apple M4 Pro · Apple M3 Max
Also runs on a Apple M3 Pro: DeepSeek-R1 (distill) · Qwen2.5 · Mistral 7B · Qwen3 · Llama 3.1
Best models for a Apple M3 Pro · 18GB →
These share Gemma 4's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M3 Pro · 18GB too:
| model | shares | verdict here |
|---|---|---|
| Gemma 3 | 8kv/256hd | ✅ Runs (tight) |
| Gemma 2 | 8kv/256hd | ✅ Runs (tight) |
Check any combo yourself: open the checker →