Qwen2.5 on Apple M4 Max · 128GB

✅ Runs comfortably — Qwen2.5 72B @ Q5_K_M
Qwen2.5 72B at Q5_K_M needs ~53.1 GB (weights 48.2 GB + KV 2.5 GB + overhead 2.4 GB @ 8K ctx) of your 93.0 GB usable (of 128 GB unified memory) — plenty of headroom. Expect ~4–7 tok/s (slow).
context @ Q5_K_M: comfortable to 64K · runs to its full 128K
quantneedsspeed
✓ Q5_K_M53.1 GB~4–7 tok/s (slow)
✓ Q4_K_M45.6 GB~5–8 tok/s (slow)
✓ Q3_K_M37.2 GB~6–10 tok/s (usable)
✓ Q2_K32.3 GB~7–12 tok/s (usable)
$runlocal install qwen2.5:72b
runlocal verdict card
download card share on x share on reddit drop it in your model-drop post
Qwen2.5: Runs comfortably
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 M4 Max: DeepSeek-R1 (distill) · Qwen3 · Llama 3.1 · Llama 3.3 · Mistral Small 3

Best models for a Apple M4 Max · 128GB →

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 M4 Max · 128GB too:

modelsharesverdict here
Qwen2.5-Coder 2kv/128hd4kv/128hd8kv/128hd ✅ Runs comfortably
Llama 3.2 8kv/128hd ✅ Runs comfortably
Mistral 7B 8kv/128hd ✅ Runs comfortably
Mistral Nemo 8kv/128hd ✅ Runs comfortably
Codestral 8kv/128hd ✅ Runs comfortably
Llama 3.2 Vision 8kv/128hd ✅ Runs comfortably
Mixtral 8x7B 8kv/128hd ✅ Runs comfortably
Phi-4-mini 8kv/128hd ✅ Runs comfortably

…and 10 more — see the Qwen2.5 page.

Check any combo yourself: open the checker →