Summary: France’s Mistral AI just opened a public preview of its new flagship model, Mistral Large 4. It has 1 trillion parameters and handles text, code, and images. The free open-weight release is planned for later in October 2026.
What happened
On October 6, 2026, French AI company Mistral announced the public preview of Mistral Large 4. The team calls it “le Chonk.” It is the company’s first big new model since December 2025.
Right now you can try the model through Mistral Studio, the company’s developer platform. It is an API preview. That means developers can connect the model to their own apps and test it. The free downloadable version is not out yet. Mistral says the model weights will come by the end of October. Reuters reported October 27 as the target date.
Why this is big news
Open models are AI models that anyone can download and run. Mistral is famous for this. When the weights drop, anyone with enough hardware can run Large 4 on their own computers. They can also change it and build on top of it. That is very different from closed models like GPT-6 Astra or Claude Opus 5.5, which you can only use through the company’s website or API.
Mistral says Large 4 is now among the best open models in the world. It also claims it is the strongest open-weight model built outside China, by a clear margin. These are company claims. Independent ranking sites have not confirmed them yet.
The model has 1 trillion total parameters, with 49 billion active at any time. A parameter is like a tiny dial the model tunes while learning. More parameters usually mean a smarter model. But only 49 billion are active for each task, which keeps running costs lower.
It is natively multimodal. That means it was built from the start to understand text, code, and images together. It can read long documents, answer questions about pictures, and write code.
How it was built
Mistral trained Large 4 from scratch on 3,800 NVIDIA Grace Blackwell GPUs. These are some of the most powerful AI chips in the world. The training happened in Mistral’s own data centers in Europe.
The training data covers more than 160 languages. That includes every official language of the European Union. Strong support for many languages matters for students and freelancers outside the English-speaking world. You are not limited to English only.
The company is also testing the model with cybersecurity experts and government partners before the full release. They want to check for risks first. Mistral says some partners get a special version with fewer limits for this testing.
The price (preview rates)
Mistral shared preview pricing for the API:
- Input: $0.68 per million tokens
- Cached input: $0.07 per million tokens
- Output: $2.09 per million tokens
These are low preview rates. The normal rates shown are higher: $1.36 input, $0.14 cached, and $4.18 output. Mistral may change these prices later. So check the live price in Mistral Studio before you build anything serious.
A token is a small piece of text, like part of a word. One million tokens is roughly 750,000 English words. So the preview price is very cheap for developers who want to build AI tools.
What it is good at
Mistral says Large 4 leads all open models in three areas: cybersecurity, finance, and legal work. It also claims that in some special tasks, like visual grounding (finding objects in images), it beats even the best closed models.
Take these claims with care. They come from Mistral’s own tests. Independent tests have not confirmed them yet. Still, even if only half of the claims hold up, this is a strong model.
The model supports many developer features: structured outputs, function calling, document question answering, batch processing, and AI agents. In simple words, you can connect it to software so it does real work, not just chat.
Why it matters for freelancers
Freelancers who build AI services can use Large 4’s cheap API to power client projects. If a client needs an AI chatbot or a document reader, you can build it on Large 4 at a low cost. The open weights also mean you can run it for clients who care about data privacy. Their data never has to leave their own servers.
Freelancers in Pakistan and other countries should watch the October weight release. A top-class open model you can run yourself can cut your tool costs to nearly zero.
For more tools you can use today, read our guide to the best AI tools for freelancers.
Why it matters for students
Students get two wins. First, the API preview is cheap, so you can build class projects and portfolio pieces without spending much. Second, once the weights are out, students and researchers can study a top-level model for free. You can see how it works inside. That is the best way to learn real AI skills.
If you are learning AI development, Large 4’s strong skills in many languages also mean you can build tools in your own language, not just English.
See our list of free AI tools for students to pair with this news.
The comeback story
Mistral needed this win. Its last big model came in December 2025. Since then, American and Chinese models moved ahead. Mistral’s models had slipped to 24th place on Artificial Analysis, a well-known independent ranking site.
The company also raised big money recently. A funding round led by Samsung pushed its value above $24 billion. It is now building its own data centers in Europe, so it does not depend on American cloud companies.
What to watch next
- The weight release (expected October 27): This is the real event. The preview is nice, but the free download is what makes this news matter for most people.
- Independent benchmarks: Watch sites like Artificial Analysis to see if Mistral’s claims hold up.
- Final pricing: Preview prices are low. The real prices may be higher.
The bottom line
Mistral Large 4 is the most important open model launch of October 2026 so far. The preview is live now. The free weights come later this month. If you are a freelancer or student who wants powerful AI without paying big-company prices, keep your eye on October 27.
Source: Mistral AI’s official announcement (October 6, 2026), with reporting by Reuters, Le Monde, and The Wall Street Journal.