Mistral Large 3 — local AI model by Mistral AI
Written by Jakub Rusinowski · Last updated
Mistral's open-weight flagship, released 2 December 2025: a 675B MoE with 41B active parameters, Apache-2.0, 256K context. Datacentre-scale to self-host; most readers will rent it.
Variants
The smallest Mistral Large 3 variant needs about 408 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Mistral Large 3 675B-A41B → 675B (41B active) | ~408.3 GB | 256,000 | ollama run mistral-large-3 |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Mistral Large 3 locally
Install Ollama, then pull the tag.
ollama run mistral-large-3Pick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: Mistral Large 3 675B-A41BRecommended GPU
The cheapest catalogued GPU that runs Mistral Large 3 locally (min 408 GB VRAM) is the Apple M3 Ultra (512 GB).
Can I run Mistral Large 3 on my GPU?
Mistral Large 3 — frequently asked questions
How much VRAM does Mistral Large 3 need?
Mistral Large 3 needs about 408 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Mistral Large 3 675B-A41B (408 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Mistral Large 3 on an RTX 4090 (24 GB)?
Mistral Large 3's smallest variant needs about 408 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for Mistral Large 3?
Q4_K_M is the best balance of quality and VRAM for Mistral Large 3 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Mistral Large 3 with Ollama?
Mistral Large 3 has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.