Mistral AI SAS has released the open weights of its frontier language model, Mistral Large 4 (ML4), also nicknamed “Le Chonk.” The model, initially previewed about three weeks earlier, is now available on Hugging Face, enabling users to run it with greater control over deployment.
Mistral positions ML4 as a highly capable, “most capable” model to date and says it is natively multimodal, with a total of 1 trillion parameters (and 52 billion active parameters, per the provided summaries). SiliconANGLE and The Next Web both emphasize the practical implications of open weights, including “sovereign” control. The New Stack highlights performance comparisons, reporting a claimed coding lead over GLM-5.3, while The Next Web adds that additional reinforcement learning guided by customer feedback contributes to improvements.
Across outlets, the core facts align: ML4 open weights have been published on Hugging Face following a preview. The differences are mainly in emphasis—sovereign control, parameter/model characterization, and performance framing versus specific competitive claims and claimed gains in coding and cyber defense.
