Mistral has released a preview of Mistral Large 4, a 1 trillion parameter model with 49 billion active parameters, trained on 3,800 NVIDIA Grace Blackwell GPUs. Open weights are promised for the end of October 2026.

Mistral Large 4 Preview

Mistral has released a preview of Mistral Large 4, a 1 trillion parameter model with 49 billion active parameters. The model was trained on the company's own cluster of 3,800 NVIDIA Grace Blackwell GPUs and is currently available through the Mistral API. Mistral has promised to release the open weights model at the end of this month.

Reasoning Levels and Output

The model supports only two reasoning levels, "none" and "high", via the Mistral API. In a test generating an image of a pelican riding a bicycle, the "high" setting produced a better result while using 2,717 output tokens, compared to 3,275 tokens for the "none" setting.

Benchmark Scores and Context

On Artificial Analysis, Mistral Large 4 scores 38, placing it just behind DeepSeek 4.1 Flash, a 552B model. This marks a significant improvement over Mistral Large 3 from December 2025, which scored 9 on the same benchmark and produced a notably worse pelican image. The new model is not considered Fable-class, but it puts Mistral back to being roughly six months behind the frontier.

Sources: Simonwillison

Mistral Large 4 Preview Arrives With 1 Trillion Parameters and Open Weights Promised