Quick Answer
- Mistral Large 4 (nicknamed “Le Chonk”) launched as a public API preview on October 6, 2026.
- It has about 1 trillion parameters, with 49 billion active at a time (mixture of experts). It handles text and images.
- API price: $1.36 per million input tokens and $4.18 per million output tokens.
- Open weights are promised for the end of October 2026. The license has not been announced.
- Mistral is releasing it in stages because of cybersecurity risks.
What is Mistral Large 4?
Mistral AI is a French AI company. Large 4 is its biggest model so far. It is a mixture of experts model, which means only a part of the model works on each request. Here, 49 billion of about 1 trillion parameters are active at once, which keeps it faster and cheaper than running the whole thing.
It is natively multimodal, so it reads both text and images, and Mistral says it supports more than 160 languages. It was trained in Mistral’s own European data centers on Nvidia Grace Blackwell GPUs, about 3,800 of them according to Mistral. Some outlets report 4,000, so we use Mistral’s own number.
Mistral Large 4 price
| Item | Price |
|---|---|
| Input | $1.36 per 1M tokens |
| Output | $4.18 per 1M tokens |
| Cached input | $0.14 per 1M tokens (one outlet, MarkTechPost) |
For a price comparison with other top models, see our Claude Opus 5.5 vs GPT-6 Sol and Luna price war guide.
Release date and open weights
- October 6, 2026: API preview available through Mistral Studio
- End of October 2026: model weights promised
Once weights are out, developers can download and run the model themselves. As of now, they are not available, and the license has not been announced.
Why is Mistral releasing it in stages?
Mistral says it wants security testing first. Open weights give users full control, but that also makes it easier to remove safety limits. So it is running a moderated public API, plus a special tier for “cybersecurity leaders, vetted partners and state authorities” with reduced moderation and expanded cyber features. This red-teaming period is how it plans to check safety before the full release.
What can it do?
These numbers come from Mistral and early coverage, not independent tests yet:
- Cybersecurity: about 93% on Cybench and 82% on a vulnerability reproduction test
- Coding: 61.7% on DeepSWE and 59.4% on SWE-Atlas-QnA
- Safety: 93.3% attack resistance on the Lakera B3 benchmark
Treat these as company claims until outside testers repeat them.
How good is it compared with others?
One outside measure, the Artificial Analysis Intelligence Index, reportedly scored it 38. It is described as the top Western open-weights result, but still behind the leading Chinese open models. You can read about one of those in our DeepSeek V4.1 Flash explainer.
Who should care?
- Developers who want a strong open model to host themselves, mainly in Europe
- Companies with data rules that prefer European infrastructure
- Security teams, as the model is built with cyber work in mind
If you only want a chatbot, nothing changes for you yet.
FAQ
When is Mistral Large 4 released?
The API preview launched on October 6, 2026. Open weights are promised for the end of October.
How much does Mistral Large 4 cost?
$1.36 per million input tokens and $4.18 per million output tokens through the API.
Is Mistral Large 4 open source?
It is described as open weight, but the weights are not out yet and the license has not been announced.
How many parameters does it have?
About 1 trillion in total, with 49 billion active at a time.
Can it understand images?
Yes, it handles text and vision.
Is it better than Chinese open models?
Not according to the Artificial Analysis score cited by one report. It leads Western open models but trails the top Chinese ones.
