Details
- Mistral AI announced Mistral Large 4, nicknamed Le Chonk, on October 6, 2026.
- The model has 1 trillion parameters, with 49 billion active parameters per request, and supports text and image inputs natively.
- It combines instruction-following, reasoning and agentic capabilities in a single mixture-of-experts model.
- Mistral says Large 4 leads aggregated open-weight benchmarks among models from the United States and Europe.
- The company reports state-of-the-art performance for cyber defense, manufacturing and finance workloads, and says it surpasses closed frontier models on visual grounding.
- Large 4 is available immediately through the Mistral API in public preview, while its model weights are planned for release at the end of October.
- Mistral describes the system as built and deployable in Europe through its own cloud infrastructure, with private work underway with cybersecurity partners.
Impact
Large 4 strengthens Europe’s position in open-weight AI and gives developers an alternative to closed systems from major US providers. Its mixture-of-experts design limits active computation to 49 billion parameters despite a trillion-parameter total, which could improve deployment efficiency, although the model’s hardware and operating costs remain unclear. The planned weights release may also support local and sovereign deployments, particularly in regulated sectors such as finance, manufacturing and government cybersecurity.
