Mistral launches Large 4, a 1 trillion parameter model, in preview

What is Mistral Large 4?

Mistral Large 4, nicknamed ML4, is the largest model from the French AI company Mistral. Mistral says it is natively multimodal, so it works with text and images, and that it was trained from scratch on 3,800 Nvidia Grace Blackwell GPUs in its own datacenters in Europe.

How big is Mistral Large 4?

Mistral says it has 1 trillion parameters, with 49 billion active at a time. Parameters are the numbers a model learns during training. More of them usually means a more capable model, but also a more expensive one to run.

Who can use Mistral Large 4 now?

Mistral calls this launch a public preview. The preview API is open on Mistral Studio. Mistral also says cybersecurity leaders, vetted partners and state authorities are red-teaming the model, using a version with reduced moderation and expanded cyber capabilities.

What do Mistral's benchmarks show?

Mistral reports that the model solves 93% of the challenges in Cybench, a set of 40 security exercises, and scores 61.7% on DeepSWE v1.1, a coding test. It says some coding scores were measured privately by Artificial Analysis. These are Mistral's own figures. We have not tested them.

ML4 pushes the frontier of open-weight performance.

Mistral

Are the weights available?

Not yet. Weights are the files that let anyone download a model and run it on their own hardware. Mistral says it will release them by the end of the month. We will update this story when they appear.

Why does it matter?

Most top models can only be used through a company's servers. If Mistral releases the weights as promised, companies and governments could run a model of this size themselves, under their own rules.

What's next?

Mistral says it will share more on the model's architecture, extra benchmarks and how it was post-trained, and release the weights by the end of October 2026.