Le Chonk — officially Mistral Large 4 — is the French company's biggest open weights release to date. The trillion-parameter model uses a mixture-of-experts architecture with roughly 49 billion active parameters, and was trained on about 3,800 Nvidia Grace Blackwell GPUs across Mistral's own European datacenters on a corpus spanning more than 160 languages. Mistral frames the release as a sovereign AI play, offering frontier-class capabilities trained and run within Europe.
Independent testing from Artificial Analysis places the preview model between DeepSeek V4.1 Flash and OpenAI's entry-level GPT6 Luna, behind the flagship models from OpenAI and Anthropic but ahead of Thinking Machines Lab's Inkling, the most capable US open weights model. Mistral's own benchmarks claim it trades blows with top Chinese offerings from Alibaba, Moonshot, Z.AI, and DeepSeek, and the company highlights strong cybersecurity performance, arguing that open weights let organizations run red-team work under their own policies rather than facing provider-level refusals.
Mistral did not publish hardware requirements, but the model is small enough to run on 8-way GPU systems such as Nvidia's HGX B300 or AMD's MI355X. Weights are expected on Hugging Face and other repositories later this month; the model is currently available via Mistral's API. Mistral says Le Chonk is the first in a series of models following its €3 billion Series D round, with further reinforcement learning expected to improve performance.