AI News · Models ·
Mistral unveils trillion-parameter Large 4, with public release set for October 27
Mistral unveiled Mistral Large 4, also called Le Chonk, a 1-trillion-parameter open-weight model. The company says it performs strongly in cybersecurity, coding, manufacturing, finance and multimodal tasks. Public availability is scheduled for October 27, with cybersecurity experts and government authorities receiving earlier access to a version with fewer safety restrictions for testing.
Key points
- Mistral Large 4 is a 1-trillion-parameter open-weight model scheduled for public availability on October 27.
- Mistral says it is the strongest open-weight model developed outside China.
- CNBC reported that Large 4 still trails the frontier in areas such as coding.
- Cybersecurity experts and government authorities receive early access to a less restricted version for testing.
- Business deployment hardware requirements, licensing terms and costs were not reported.
What happened: Mistral has unveiled Mistral Large 4, also known as Le Chonk, a 1-trillion-parameter open-weight AI model, with public availability scheduled for October 27. The French company says the model is particularly effective in cybersecurity, coding, manufacturing, finance and multimodal tasks. CNBC reported that Mistral described it as the strongest open-weight model developed outside China by a substantial margin. Those performance assessments are company claims, rather than independent findings reported by CNBC.
The details: Mistral trained Large 4 on 4,000 Nvidia Grace Blackwell GPUs over two months in its own European data centers, CNBC reported. The company says the model will rank among the leading open-weight systems globally on aggregate benchmark performance when its core parameters are released. However, CNBC also reported that it still trails the frontier in areas such as coding. That distinction matters for business buyers: a strong overall ranking does not establish that a model is the best fit for a particular workflow. Detailed benchmark scores and business deployment requirements were not reported.
Background: Open models can be modified and self-hosted, unlike the leading closed systems from OpenAI and Anthropic. CNBC described access to these models as a growing flashpoint in technological competition between the U.S. and China, with Chinese open models increasingly adopted worldwide. Large 4 enters that market as Western developers seek to close the capability gap. Mistral raised 3 billion euros in a September Series D round at a 21 billion euro valuation. Co-founder and chief scientist Guillaume Lample said the company expects model capabilities to improve as it expands training capacity following that fundraising.
Who it affects: For teams seeking a model they can modify or run themselves, Large 4 could add another option. Its early rollout also puts cybersecurity users at the center of the launch. Developers, cybersecurity leaders and government authorities are receiving preview access before the wider release. Cybersecurity experts and authorities are getting earlier access to a version with fewer safety restrictions for testing. Lample said the model's cyber defense capabilities would help enterprises and governments defend against attackers using jailbroken closed models.
What to watch: The October 27 release should give prospective users an opportunity to assess Mistral's claims against their own requirements. Buyers should distinguish the resources used to train Large 4 from what would be needed to deploy it: hardware requirements for business use were not reported. Licensing terms, deployment costs and detailed independent evaluations were also not reported. Until those details are clear, the announcement establishes a new contender rather than a proven fit for any particular enterprise workload.
Our take
Large 4 could expand the options for teams seeking self-hosted frontier models. Buyers should validate Mistral's performance claims and deployment requirements before committing.