Mistral previews Large 4 and says it will publish the weights by the end of October

On October 6, 2026, the French company Mistral AI opened a public preview of Mistral Large 4 through its API on Mistral Studio. It is the company’s largest model so far: about 1 trillion parameters, 52 billion of them active at a time, and it reads images, documents and charts as well as text. Mistral lists prices of $1.36 per million input tokens and $4.18 per million output tokens (a token is roughly a piece of a word).

Mistral says it will publish the model’s weights, the files that make up the trained model, by the end of October. TechCrunch reported that the release will follow safety testing. Neither source names a license. The model was trained in Mistral’s own datacenters in Europe, and the preview runs on that same infrastructure. Mistral also plans a European deployment that it operates end to end under European law, and will offer the model in other regions. The company chose which benchmark scores to publish, from its own tests and from outside evaluators.

What it means for your organization

Once the weights are out, an organization could run a large model on its own servers or private cloud, or through a provider it chooses. That matters for anyone with strict confidentiality or data-location requirements, and it adds an option outside the large U.S. providers.

What to check

For now the weights are only a promise. Until they come out with a license, nobody can say whether commercial use will be allowed, or on what terms. A model this size also needs serious hardware, so most organizations would use a hosted version. If you do, ask the provider where the model runs and how long it keeps your data. And test it on your own tasks before trusting the published scores.

Sources

• Mistral AI, October 6, 2026: Mistral Large 4 announcement
• TechCrunch, October 6, 2026: Mistral’s new 1T model aims to leapfrog closed and open rivals