French AI startup Mistral rolled out its Mistral Large 4 model on Tuesday, claiming it can outperform other U.S. and EU open-weight models.
Mistral Large 4, nicknamed "le Chonk," can manage tasks like code writing, processing images and text, and using software to carry out more complex work.
Mistral's public preview stat sheet says the model has over 1 trillion total parameters.
Model 4 performance
Mistral says Large 4 is on par with or better than the strongest open models on the market.
In the "Harvey's Legal Agent" benchmark, which tests a model's ability to handle complex legal tasks, Large 4 scored nearly 3% higher than the next best model, Kimi K3, and more than 10% higher than OpenAI's GPT-6 Astra.
ML4 was also competitive with other major models on coding and business workflows.
Its cybersecurity benchmark scores were also a standout. Large 4 scored 50 on the Artificial Analysis Cyber Index, measuring how well AI finds software security flaws and fixes them. It tied the other top model, GLM-5.3-Flash, and finished ahead of DeepSeek-V4.1-Flash and Kimi K3.
The company billed Model 4 as part of its push for AI sovereignty, saying it is "designed to give customers control over their AI."
Mistral expects weights due by the end of October, at which point users will be able to deploy Large 4 on their own infrastructure.