Qwen2.5 on Apple M3 Max · 36GB

✅ Runs (tight) — Qwen2.5 32B @ Q3_K_M
Qwen2.5 32B at Q3_K_M needs ~17.7 GB (weights 14.9 GB + KV 2.0 GB + overhead 764 MB @ 8K ctx) of your 21.1 GB usable (of 36.0 GB unified memory) — fits, but little headroom — close other apps or trim context. Expect ~10–16 tok/s (usable).
context @ Q3_K_M: comfortable to 4K · runs to 64K · won't fit past that on this rig
quantneedsspeed
✗ Q8_036.0 GB—
~ Q5_K_M24.8 GB~0.9–1 tok/s (slow)
~ Q4_K_M21.4 GB~1–2 tok/s (slow)
✓ Q3_K_M17.7 GB~10–16 tok/s (usable)
$runlocal install qwen2.5:32b
runlocal verdict card
download card share on x share on reddit drop it in your model-drop post
Qwen2.5: Runs (tight)
for your own README — links back here

Same model, other machines: Apple M1 · Apple M1 Pro · Apple M2 · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro

Also runs on a Apple M3 Max: Qwen3 · DeepSeek-R1 (distill) · Gemma 4 · Gemma 3 · Mistral Small 3

Best models for a Apple M3 Max · 36GB →

Architecturally similar

These share Qwen2.5's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M3 Max · 36GB too:

modelsharesverdict here
Llama 3.1 8kv/128hd ✅ Runs comfortably
Qwen2.5-Coder 2kv/128hd4kv/128hd8kv/128hd ✅ Runs (tight)
Llama 3.2 8kv/128hd ✅ Runs comfortably
Llama 3.3 8kv/128hd 🐢 CPU-only (slow)
Mistral 7B 8kv/128hd ✅ Runs comfortably
Mistral Nemo 8kv/128hd ✅ Runs comfortably
Codestral 8kv/128hd ✅ Runs (tight)
Llama 3.2 Vision 8kv/128hd ✅ Runs comfortably

…and 12 more — see the Qwen2.5 page.

Check any combo yourself: open the checker →