Ministral 3 14B on Apple M1 Pro · 16GB
🐢 CPU-only (slow) — Ministral 3 14B @ Q4_K_M
Ministral 3 14B 14B at Q4_K_M needs ~9.4 GB (weights 7.7 GB + KV 1.3 GB + overhead 512 MB @ 8K ctx) of your 8.2 GB usable (of 16.0 GB unified memory) — runs on CPU only. Expect ~2–3 tok/s (slow).
context @ Q4_K_M: runs to 32K · won't fit past that on this rig
| quant | needs | speed |
|---|---|---|
| ✗ Q8_0 | 15.3 GB | — |
| ~ Q4_K_M | 9.4 GB | ~2–3 tok/s (slow) |
for your own README — links back here
Same model, other machines: Apple M1 · Apple M2 · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro · Apple M3 Max
Also runs on a Apple M1 Pro: DeepSeek-R1 (distill) · Qwen3 · Llama 3.1 · Qwen2.5 · Mistral 7B
Best models for a Apple M1 Pro · 16GB →
Architecturally similar
These share Ministral 3 14B's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M1 Pro · 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 Small 3 | 8kv/128hd | 🐢 CPU-only (slow) |
| Mistral Nemo | 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 Ministral 3 14B page.
Check any combo yourself: open the checker →