Mistral launches Large 4, a 1-trillion-parameter model

Anyone who wants to try Mistral Large 4 can do so now, but only through a paid API: the French company Mistral AI opened a preview on Mistral Studio, its developer platform, on Oct. 6. It costs $1.36 per million input tokens and $4.18 per million output tokens (a token is a fragment of a word). There is no separate pricing or availability announcement for Mexico.

For ordinary users nothing changes yet. For developers, the news comes later: Mistral says it will publish the weights, the files that let anyone run the model on their own servers, "at the end of October." TechCrunch reports a window of up to three weeks, contingent on safety testing. The license has not been detailed. "Open weights" does not always mean open source: the license decides that.

What the company says

According to Mistral, the model has 1 trillion parameters, of which only 49 billion activate on each query (a mixture-of-experts design). It accepts text and images and handles more than 160 languages. It was trained on 3 800 NVIDIA Grace Blackwell GPUs; TechCrunch cites about 4 000 from an executive, and the gap has not been explained.

These are the performance figures published by the company itself:

• 61.7% on DeepSWE 1.1 and 28.3% on Terminal Bench 4.0.
• 93% on Cybench, a vulnerability-patching test.
• 59.9% on AutomationBench and 42% on Dense 200.

No third party has measured any of them. Mistral also describes the model as "state-of-the-art" among open-weight models developed in the U.S. or Europe, a claim with no published independent comparison.

Pierre Stock, Mistral's vice president of science, told TechCrunch that training used two to three times less compute than Chinese competitors and significantly less than closed-source rivals. No comparable figures back that up.

Its predecessor, Mistral Large 3, appeared in December 2025 with 675 billion parameters. Outside evaluations and the Large 4 license will show how much of the announcement holds up once the weights are in other people's hands.