Run Mistral Large 4 on M3 Ultra

Mistral Large 4 runs at 15 tok/s on a 512 GB M3 Ultra, a published figure reported by huggingface.co (llama.cpp, Q2_K, prompt length not stated (prefill ~90 tok/s)). It needs about 383 GB of unified memory. It is Mistral's 1.05T mixture-of-experts model (49B active), released under the Mistral Ecosystem Open-Weight licence.

Speed
sourced15
tok/s
First Token
—
seconds
RAM Needed
383
GB minimum
Engine
llama.cpp
of this figure
Want to run Mistral Large 4 faster — or step up to bigger models? Find your Mac →

The figure

This is a published figure: a real run, reported by its source and linked below. The index records it as given and never adjusts it.

MetricValue
Tokens per secondsourced15 tok/s
Provenancepublished figure · huggingface.co
QuantizationQ2_K
Enginellama.cpp
Contextprompt length not stated (prefill ~90 tok/s)
Figure dated2026-10-29
Minimum RAM383 GB
Parameters1.05T MoE (49B active)
Chip bandwidth819 GB/s

Every figure for Mistral Large 4 on the M3 Ultra

Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (383 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.

SpeedQuantEngineMac RAMContextSourceNote
sourced15 tok/sQ2_Kllama.cpp512 GBprompt length not stated (prefill ~90 tok/s)huggingface.coHugging Face launch post: llama serve -hf ggml-org/Mistral-Large-4-1T-A52B-GGUF:Q2_K (380.7 GB, ~2.9 bpw) through llama.cpp on one 512 GB M3 Ultra; build not stated (support merged 29 Oct, first release b11998).
Provenance key
sourced
Published figure — vendor or press review
estimated
community
Linked individual run — blog, forum or submission

Setup: run Mistral Large 4 on a Mac

Ollama — the index has not verified an Ollama tag for Mistral Large 4, so it doesn’t print one (a guessed tag fails with “pull model manifest: file does not exist”). Search the Ollama library →

MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:

pip install -U mlx-lm
mlx_lm.generate --model <repo-id> --prompt "Hello"

Downloads: Hugging Face repo → · Hugging Face model card →

The 15 tok/s figure above came from llama.cpp at Q2_K; other runtimes and quants run at different speeds.

Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.

Mistral Large 4 on other Apple Silicon chips

ChipSpeedQuantEngineBandwidth
M5 Ultraestimated21 tok/s3.0 bpwllama.cpp1228 GB/s

For the estimated figures here, the bandwidth formula’s typical error for large MoEs is ±34% (mostly below the measured figure), against 70 measured runs of 18 models (how it is computed).

About Mistral Large 4

Mistral's ~1T MoE (49B active, 52B with embeddings), weights 29 Oct 2026: 1M context, image input, under the new Mistral Ecosystem Open-Weight License (an MIT-style grant, but Model-as-a-Service operators need a separate commercial licence; internal use and products that embed the model are exempt). Q2_K is 381 GB, so it needs a 512 GB Mac Studio: Hugging Face's launch post reports ~15 tok/s generation on a 512 GB M3 Ultra with llama.cpp. llama.cpp supports it from b11998; stock mlx-lm does not yet (PR #2018) but mlx-vlm 0.7.12 does, and Ollama is cloud-only.

Mistral Ecosystem Open-Weight licence · 1.05T MoE · 1000K context · LLMCheck capability 45/50 · see its leaderboard rank →

Memory by context length

Estimated, not measured: the index’s minimum for Mistral Large 4 (383 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. MLA (512+64); 9 of 72 layers reuse another layer's latent (kv_source_layers), so 63 store one; indexer keys not counted. A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M3 Ultra Mac on macOS 15, the lower default; macOS 26 allows more): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).

ContextKV cache (16-bit)RAM neededRAM, q8 KVSmallest tier that fitsWithin the default GPU limit
8K~0.6 GB~383.3 GB~383.1 GB512 GB Macsame
32K~2.4 GB~385.1 GB~384.0 GB512 GB Mac raise GPU limit (macOS 15)more than any Mac’s default
64K (coding-agent context)~4.8 GB~387.5 GB~385.2 GB512 GB Mac raise GPU limit (macOS 15)more than any Mac’s default
128K~9.5 GB~392.2 GB~387.6 GB512 GB Mac raise GPU limit (macOS 15)more than any Mac’s default

System requirements

Macs with the M3 Ultra

🛒 Get a Mac that runs Mistral Large 4

Mistral Large 4 needs about 383 GB of unified memory. These Apple Silicon Macs have enough to run it. Apple builds the higher-memory configurations to order; Amazon does not stock them, so the used option links Amazon Renewed listings.

As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.

Compare More Models

See how Mistral Large 4 stacks up against other models on your specific Mac hardware.

Open Compare Tool Full Leaderboard