Llama 3.3 on Apple M4 Max · 128GB

✅ Runs (tight) — Llama 3.3 70B @ Q8_0
Llama 3.3 70B at Q8_0 needs ~75.8 GB (weights 69.8 GB + KV 2.5 GB + overhead 3.5 GB @ 8K ctx) of your 93.0 GB usable (of 128 GB unified memory) — fits, but little headroom — close other apps or trim context. Expect ~3–5 tok/s (slow).
context @ Q8_0: comfortable to 2K · runs to its full 128K
quantneedsspeed
✓ Q8_075.8 GB~3–5 tok/s (slow)
✓ Q5_K_M51.3 GB~5–8 tok/s (slow)
✓ Q4_K_M44.1 GB~5–9 tok/s (slow)
✓ Q3_K_M36.0 GB~6–11 tok/s (usable)
✓ Q2_K28.3 GB~8–14 tok/s (usable)
$runlocal install llama3.3:70b
runlocal verdict card
download card share on x share on reddit drop it in your model-drop post
Llama 3.3: 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 M4 Max: DeepSeek-R1 (distill) · Qwen3 · Qwen2.5 · Llama 3.1 · Mistral Small 3

Best models for a Apple M4 Max · 128GB →

Architecturally similar

These share Llama 3.3'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 8kv/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 6 more — see the Llama 3.3 page.

Check any combo yourself: open the checker →