TL;DR
What is Mistral Large 4 and should you care?
Mistral released its biggest AI model, Large 4, for anyone to download and use, mostly for free. It's strong at coding and cheap to use through apps. Most small businesses don't need to do anything yet, but it should help keep AI prices down.
What did Mistral release?
On October 29, 2026, Mistral released the weights of an improved Mistral Large 4 on Hugging Face, the main site for sharing AI models. Mistral nicknames it "Le Chonk," and with 1 trillion parameters, the name is earned. A preview came out earlier this month.
Mistral says it's now the best open-weight model for coding and strong at defensive cybersecurity, meaning helping protect systems. It handles images as well as text, and it has three reasoning modes (low, medium and high) so you can trade speed and cost for quality.
Quick terms
- Weights: the trained "brain" of an AI model. Open weights means anyone can download and run it.
- Parameters: the settings inside the model. More usually means more capable, and harder to run.
- Tokens: small chunks of text, roughly three-quarters of a word. AI services charge per token.
- Multimodal: the model reads images as well as text.
How much does Mistral Large 4 cost?
The weights are free to download. If you'd rather use it through Mistral's API (the paid connection apps use), Mistral lists these prices:
| What you pay for | Price per million tokens |
|---|---|
| Input (what you send) | $0.70 |
| Cached input (repeated text it has seen recently) | $0.07 |
| Output (what it writes back) | $2.10 |
It's also available through OpenRouter and Microsoft Foundry, two services that let developers pick between many models.
What's the catch in the Mistral Large 4 license?
It uses a modified MIT license. You can use, change and sell things built with it for free. The exception: companies that want to sell Large 4 itself as a hosted model service need a separate license from Mistral. For a small business building a chatbot or tool on top of it, that exception almost certainly doesn't apply.
How good is Mistral Large 4 at coding?
These are Mistral's own picks of tests, so read them as a sales pitch with numbers:
- 48% on the Artificial Analysis Coding Agent Index, roughly tied with GLM-5.3 while using 47% fewer tokens per task. Fewer tokens means a lower bill for the same job.
- 19.7% on Cognition's FrontierCode 2.0, the top score among open models.
- 96.6% refusal rate on malicious cyber requests.
Mistral also says it saw "potential for uplift" for real attacks, meaning the model could help a bad actor, and that it's tightening guardrails.
Our take
For most readers, this release matters through their wallet. A strong free model puts pressure on every paid AI company to keep prices fair, and that tends to show up as cheaper plans and smarter free tiers. Nobody running a small shop should download a trillion-parameter model; you'd need serious hardware just to start it. Where it gets interesting is if you have a developer who builds tools for you, or if you care where your data goes. Then a model you can run on servers you choose is worth a test on one real task. We'd wait for outside test results before switching anything that works today.
— Artificially Nerdy
What to do
- Nothing, if you just use ChatGPT, Claude or Gemini and you're happy.
- Ask your developer to test Large 4 on one real coding or support task through OpenRouter and compare the cost.
- Check the license with whoever builds your tools if you plan to resell AI features.
- Keep sensitive work on tools you trust until you see results from people other than Mistral.
FAQ
Is Mistral Large 4 free?
The weights are free to download and use commercially. Using it through Mistral's API costs $0.70 per million input tokens and $2.10 per million output tokens.
Can I run Mistral Large 4 on my laptop?
No. A model this size needs powerful server hardware. Most people will use it through an app or an API.
Why is it called Le Chonk?
It's Mistral's own nickname for the model, a nod to its size.
Is Mistral Large 4 better than ChatGPT?
Mistral didn't claim that. It says Large 4 is the best open-weight model for coding, based on tests it chose.
Sources
- Mistral: Mistral Large 4 open weights announcement — October 29, 2026; license, modes, benchmarks, safety notes.
- Mistral docs: Mistral Large 4 — API prices and availability.
- Hugging Face blog: Mistral Large 4 — weights release.
- TechCrunch: Mistral releases weights for its trillion-parameter AI model targeting cybersecurity — October 29, 2026.
- The Verge: Mistral AI has officially released its Large 4 Le Chonk model — October 29, 2026.
- Mistral AI on X — October 29, 2026, 10:21am ET.
