Mistral has introduced its new flagship model, Large 4, and opened preview access to selected groups. The company highlights coding and cybersecurity among its strengths, alongside manufacturing and finance tasks. Chief scientist Guillaume Lample says the model is also intended to help businesses and governments defend against cyberattacks.
Coding, multimodal tasks and cyberdefense
Mistral says ML4 is particularly effective at coding and cybersecurity tasks. It also identifies manufacturing and finance as areas of use and highlights the model’s capabilities in multimodal tasks.
Mistral co-founder and chief scientist Guillaume Lample says attackers are bypassing safeguards in closed models and using them to carry out cyberattacks. He claims Large 4 will enable businesses and governments to defend themselves against those attackers.
One trillion parameters and open-weight comparisons
Large 4, also known as ML4 or le Chonk, has one trillion parameters, according to Mistral. Training took two months using 4,000 Nvidia Grace Blackwell GPUs in the company’s own European data centers.
Mistral claims that once the model’s weights are released, its aggregate benchmark performance will place it among the world’s leading open-weight models. The company also describes it as the strongest such model developed outside China, by a substantial margin. CNBC reports, however, that it still trails frontier models in some areas, including coding.
Broader access planned for October
Current preview access is intended for selected developers, cybersecurity representatives and state authorities. Mistral plans to make the model more widely available during October 2026.
