Mistral Unveils 1-Trillion-Parameter Large 4 AI Model

Summary

Mistral has unveiled Mistral Large 4, also called “le Chonk,” a 1-trillion-parameter flagship model that the French company says ranks among the world’s strongest open models. Mistral says the model is especially effective for cybersecurity, coding, manufacturing, finance and multimodal work, while acknowledging that it still trails frontier systems in areas such as coding. The model was trained for two months on 4,000 Nvidia Grace Blackwell GPUs in Mistral’s European data centers. Mistral says Large 4 will rank among the top open-weight models globally on aggregate benchmarks and is substantially stronger than any open-weight model developed outside China, although those claims come from the company. The model is initially available in preview to developers, cybersecurity leaders and state authorities, with a broader release planned for later in October 2026. Its core parameters are expected to be released as part of that open-weight rollout. Mistral presented the launch against a widening competition over open AI models, which can be modified and self-hosted, and the growing adoption of Chinese open models worldwide. Co-founder and chief scientist Guillaume Lample said future training capacity, supported by Mistral’s recent 3-billion-euro Series D, should improve the model’s capabilities and cyber-defense performance.