Skip to content
by InfinyAI ES

IA en un minutoNewsModels

ModelsUpdated:

Mistral unveils Large 4, a one trillion-parameter multimodal model that is in public preview for now

IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds

Large 4, which Mistral itself nicknames le Chonk, is multimodal and has a trillion parameters, with 52 billion active. For now it can only be tried through the API as a public preview, and downloadable weights are due before the month ends. It was trained from scratch in Europe, in the company's own data centers, on 3,800 NVIDIA GPUs.

Mistral has presented Mistral Large 4, the biggest model it has built so far, which the company itself has nicknamed "le Chonk".

It is multimodal and counts one trillion parameters, 52 billion of them active. At this stage it is in public preview and open to testing via API, and the open weights (the downloadable version of the model) will follow at the end of the month. Mistral is also putting the model through security testing in real-world conditions, working with state authorities, verified partners and cybersecurity leaders.

According to Mistral, Large 4 already rivals the best open models anywhere and clearly beats every open-weight model developed in Europe or the US. The company says it built the model from scratch, using 3,800 NVIDIA GPUs installed in data centers it owns in Europe. A substantial part of the training data was multilingual and covered more than 160 languages.

The preview is offered through the API on Mistral Studio, and the model card on Mistral's website puts the price at $1.36 per million input tokens and $4.18 per million output tokens. The model will be offered in several regions around the world, one of them a European deployment run from start to finish by Mistral, without relying on other digital providers and subject to European law. Every official language of the European Union is among the training languages.

On cybersecurity, Mistral says the model places in the global top five on the Artificial Analysis Cyber Index, which is an independent evaluation. When the weights are released, the company will also publish further detail on the architecture, new benchmark results and its post-training method. Large 4 is the first model on the roadmap financed by Mistral's €3 billion Series D round.

Correction (12 October 2026): the model has 52 billion active parameters, not 49 billion as this story said.

Why it matters · analysis and opinion

What changes with Large 4 is not just its size, but where the model can live. With open weights it can be installed on your own servers, and Mistral also serves it from its own European data centers, which matters to public bodies, banks or cybersecurity firms that cannot send their data elsewhere. For most, downloading it will be impractical because of the hardware it demands; the real route is the API. The comparisons are Mistral's own, so the time to judge it will come with the published weights and with third-party tests.

Official source: Mistral AI · Written with the help of AI: how we make the news

AI in your inbox, every day or every Friday

The stories that matter, each one in a minute. With the source for every one.

Choose one or both:

Sign up and you are in: the daily arrives every night and the weekly on Friday mornings. You can unsubscribe from any email.