Introducing Mistral Large 4: Le chonk
Introducing Mistral Large 4: Le chonk Mistral are back in the game. Today they're releasing a preview of Mistral Large 4, a 1 trillion parameter, 49 billion active parameter model trained on their own cluster of 3,800 NVIDIA Grace Blackwell GPUs. The preview is available via their API. They promise to release the open weights model at the "end of this month". The model only supports two reasoning…
Mistral has unveiled an impressive preview of their latest language model, Mistral Large 4. This model boasts a staggering 1 trillion parameters, with 49 billion active parameters, trained on a massive cluster of 3,800 NVIDIA Grace Blackwell GPUs. The preview is accessible through Mistral's API, with a release of the open weights model anticipated at the end of the month.
The two reasoning levels supported by the model are "none" and "high," both of which are available via the Mistral API. Among the two demonstration pelicans, the "high" version appears to be the superior choice, despite using fewer output tokens. On Artificial Analysis, the "high" pelican achieved a score of 38, closely following DeepSeek 4.1 Flash, a 552B model. This marks a significant improvement over Mistral Large 3 from December, which garnered a poor score of 9 on the same evaluation.
While Mistral Large 4 is not yet a Fable-class model, it's commendable to witness Mistral regaining its footing, now roughly six months behind the cutting edge.
Written by urgent.news from Simon Willison's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
