Le Chat just got bigger. Much, much bigger. Le Chonk has landed in preview. The open-weights will be released by the end of October 2026, after Mistral finishes red-teaming with cybersecurity partners, vetted partners, and state authorities.
We all love freedoms. In some legal justifications, the big digital firms get all the freedoms they want; the average Joanna, not so much. So those pesky Parisians at Mistral.AI have been at it again, complying with the EU AI Act and GDPR, which makes it clear: if they harmed you, they harmed you. Which is totally unfair to Big Tech, since "rogue AI" is clearly inevitable. If you buy a car where the gas tank explodes, that is a "rogue car" problem. AI is just at the cutting edge of the corporate welfare game: keeping all the profits while socialising all the losses.
Mistral even has the Charles de Gaulle to run the slogan "Mistral is #1 in Sovereign AI". Whatever that means.
Luckily, you can totally avoid this new open model unless you actively choose to use it with OpenCode AIs amazing harness or inference service. Even better, it is not yet configured by default on Mistral's own open-source Python harness, 'vibe'.
So whatever you do, DO NOT use my gist "Running Mistral Large 4 on stock Mistral Vibe — no fork required" https://gist.github.com/simbo1905/6f46cc1209ce57d9a775797ad870874b; otherwise, this could happen to your computer:
Also, never use the vibe --experimental-harness or the Python will load the new Rust-based engine to be more efficient and more secure. And don't try vibe --yolo unless you know what you are doing.
One small thing; it does feel slow just now. Reallly slowww. Yet Mistral.AI also host GLM-5.3 and of course their own models such as mistral-medium et al so you can swap between Le Chonk and Le Chat as you like.
Enjoy!
Top comments (1)
love it! But you have a lot of Charles D, to use so much reverse psychology!