Mistral has released its flagship model Large 4, codenamed Le Chonk

The French company Mistral has introduced its new flagship model Large 4, known in the community by the codename Le Chonk. It is a massive neural network with a trillion parameters that was trained entirely within Europe.

The developers claim that the model demonstrates superiority over open-source counterparts in programming tasks. According to independent testing results, Le Chonk is characterized by high speed and relatively low operating costs. Nevertheless, in general benchmarks, the model still lags behind leaders from China: it scored 26 points, while DeepSeek V4.1 Flash and GLM-5.3 show a result of 32 points.

The release of the model weights allows developers to integrate the solution into their own infrastructure projects that require high performance in code generation. For practitioners, this means the emergence of an alternative to top proprietary solutions, focused on European development standards and available for deployment within their own computing infrastructure.