Meet
Le Chonk AI.Mistral Large 4, explained.
Mistral’s biggest model.
Your practical guide to using it.
Access, costs, real-world tasks and the open-weight release. With sources, not just headlines.
What’s available?
Try it through Mistral Studio.
No checkpoint verified yet.
Commercial rights need confirmation.
Choose your next step
Try Le Chonk
Find the official Studio, choose the right model and run your first prompt.
Read guideCostsKnow what it costs
Compare standard and launch rates, then estimate your own usage.
Read guidePerformanceLook beyond the headline
Separate vendor claims from independent evaluations and actual task results.
Read guideAvailabilityCheck the weight release
See what is available today, and what is still a release plan.
Read guideA large model, a smaller active slice
Mistral Large 4, also called ML4 or Le Chonk, is a general-purpose mixture-of-experts model. It takes text and images as input and produces text. Only a subset of parameters participates in each token’s computation; that does not make the full weight storage disappear.
| Detail | Research snapshot |
|---|---|
| Total parameters | 1.05 trillion in documentation; announcement says 1 trillion |
| Active parameters | 52 billion in documentation; announcement says 49 billion |
| Context window | 1 million tokens in documentation; launch evaluation used 512k tokens |
| Model identifier | mistral-large-4 |
See Mistral Large 4 model documentation and Mistral release announcement. Differences are kept visible in the specification guide.
What people are using it for
Code & agents
Repository understanding, bug fixes and tool-assisted development.
Read guideKnowledge workDocuments & vision
Extract evidence from PDFs, inspect charts and understand images.
Read guideSecurity teamsCyber defense
Review code, verify fixes and evaluate defensive workflows.
Read guideThe October release snapshot
Sources & verification
Research snapshot: . Source statements are dated; API access and prices can change.
- Mistral release announcement
- Mistral Large 4 model documentation
- Artificial Analysis launch evaluation
Original model runs are marked “not run” unless a trace is provided. Read our editorial method.