Will Mistral-Large-3-675B-Instruct-2512 fit on a 512GB Mac Studio (M3 Ultra)?

Yes. Mistral-Large-3-675B-Instruct-2512-mlx-4bit (360 GiB) fits in one 512GB M3 Ultra Mac Studio.

mistralai/Mistral-Large-3-675B-Instruct-2512: parameter count not published in safetensors metadata. License: apache-2.0.

MLX weights	Size	On 512GB (~469 GiB GPU limit)
mlx-community/Mistral-Large-3-675B-Instruct-2512-mlx-4bit	360.2 GiB	Fits, ~109 GiB left for context

Sizes: sum of .safetensors files from the Hugging Face API (Oct 2026). "GPU limit" is iogpu.wired_limit_mb=480000, already set on our Studios. Newer architectures may need a recent mlx-lm or llama.cpp; check before you book around one model.

Run it

uv venv ~/mlx && uv pip install --python ~/mlx/bin/python -U mlx-lm
~/mlx/bin/mlx_lm.chat --model mlx-community/Mistral-Large-3-675B-Instruct-2512-mlx-4bit

Try it on a whole 512GB M3 Ultra Mac Studio

3 days $299 · 7 days $489 · 14 days $779. Bare metal, one renter per machine, mlx-lm and llama.cpp ready.

Check availability & book