Mistral Unveils Massive New AI Model, Says ‘Le Chonk’ Can Take On China’s Best Open Systems

French artificial intelligence startup Mistral has unveiled its biggest AI model yet, escalating the global competition over open AI as developers try to narrow China‘s lead in the increasingly relevant market.

The Paris-based company on Tuesday introduced Mistral Large 4, or ML4, a 1-trillion-parameter multimodal model nicknamed “le Chonk.” Mistral said the model ranks among the world’s strongest open-weight AI systems and is particularly capable in cybersecurity, coding, manufacturing, and finance.

Mistral’s own documentation lists ML4 at 1.05 trillion total parameters, with 49 billion active parameters and a 1.6-billion-parameter vision encoder. It uses a mixture-of-experts architecture, which activates only a portion of its total parameters for a given task rather than deploying the entire model at once.

The launch puts Mistral more directly into a race that has increasingly been dominated by Chinese AI developers. Companies including Z.ai, DeepSeek and Moonshot AI have helped turn China into a major source of powerful open models, which can be downloaded, modified and deployed on companies’ own infrastructure instead of being accessed exclusively through a provider’s cloud service.

Mistral claimed that ML4 is the strongest open-weight model developed outside China by a substantial margin and is competitive with some of the best Chinese systems. Early benchmark results suggest it represents a significant improvement over previous Mistral models, although independent testing will ultimately determine how closely it competes with China’s newest offerings.

The model also continues to trail leading frontier systems in some areas, including portions of coding performance. ML4 was trained over roughly two months using thousands of Nvidia Grace Blackwell GPUs deployed in Mistral’s own European data centers.

The model is initially being made available in preview, including to developers and cybersecurity specialists, with its open weights expected to be released more broadly later this month. Mistral’s documentation currently lists Large 4 as an open public-preview model with a context window of up to 1 million tokens.

“The model capabilities will further improve as we scale up our training capacity, following our Series D fundraise,” Mistral co-founder and chief scientist Guillaume Lample said. The company raised €3 billion in September at a post-money valuation of more than €21 billion. Mistral has pitched its strategy around “sovereign” and open-weight AI, allowing governments and companies greater control over where models and data are hosted.

Mistral argues that open models can be especially valuable for defensive cyber operations because security teams sometimes encounter restrictions when using closed models whose safety systems interpret legitimate penetration testing or defensive actions as malicious activity.

“Further, the cyber defense capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models to perform cyber attacks,” Lample said.