🤖Mistral Large 4: 1T Parameters, 49B Active, Open Weights Due
TL;DR
Mistral Large 4 is a 1-trillion-parameter mixture-of-experts model with 49B active, trained on 3,800 Nvidia Grace Blackwell GPUs. Simon Willison says Mistral is back in the game, roughly six months behind the frontier.
Mistral Large 4 is a 1-trillion-parameter mixture-of-experts model with 49B active, trained on 3,800 Nvidia Grace Blackwell GPUs. Simon Willison says Mistral is back in the game, roughly six months behind the frontier.
Key Points
1T total parameters, 49B active, trained on Mistral's own cluster of 3,800 Grace Blackwell GPUs
Scores 38 on Artificial Analysis versus 9 for Mistral Large 3, just behind DeepSeek 4.1 Flash (552B)
API offers only "none" and "high" reasoning levels; the pelican test used 2,717 output tokens on high and 3,275 on none
Open weights are promised by the end of October, license terms not yet stated
Why It Matters
A European lab is credible again at the near-frontier tier, and the open-weights promise is the part to watch. Without license terms, enterprises can't plan around it yet.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,566 builders reading daily.