French company Mistral has released Mistral Large 4, a 1-trillion-parameter model nicknamed 'Le Chonk,' which it claims can compete with leading proprietary models from the US and China, according to WIRED and Ars Technica. The model is available in preview, with a final version expected by the end of the month. It is open-weight, meaning anyone can use and customize it, and is optimized for coding and cyberdefense as well as tasks in manufacturing, finance, and electrical engineering. Mistral presents it as the most capable open-weight model developed outside China and 'very, very close' to some proprietary models, and says it was trained from scratch rather than through distillation. Cofounder and chief scientist Guillaume Lample told WIRED, 'Mistral is still in the race of getting the best model. This is the main message.' The reports note Mistral has generally lagged US labs on performance, revenue, and release frequency but recently raised a $3.3 billion funding round at a $24 billion valuation in September, which WIRED describes as the largest ever by a European tech company. The release coincides with US restrictions, placed in June, on distributing OpenAI and Anthropic models, and broader debate over access to frontier AI.
- Mistral Large 4 ('Le Chonk') is a 1-trillion-parameter open-weight model, in preview with a final version due by month's end
- It is optimized for coding, cyberdefense, manufacturing, finance, and electrical engineering, and Mistral says it was trained from scratch
- Mistral raised $3.3 billion at a $24 billion valuation in September, per WIRED
What it means for you
There's a new freely available AI model from a European company that you can run yourself or through Mistral's cloud, rather than renting access from OpenAI or Anthropic. The pitch is cost and control: open-weight models cost only the compute they use, and no vendor can revoke your access overnight. Mistral's own performance claims ('very, very close' to proprietary models) are unverified by independent benchmarks, so treat them as marketing until tested.
Try this
If you already run or evaluate open-weight models, add Mistral Large 4 to your next comparison test against whatever you use now, using your own real tasks rather than the vendor's claims.
Who should care
Technical teams and businesses that self-host models, worry about vendor lock-in, or need a European-based provider for data-sovereignty reasons—especially those doing coding or cyberdefense work.
Skip this if
You use ChatGPT, Claude, or Gemini through a normal subscription and have no plans to self-host a model—this changes nothing for you today.
Sources: Ars Technica AI, Wired AI — read the original