Mistral has previewed its newest large AI model, Mistral Large 4—nicknamed “Le Chonk.” Multiple outlets report the model is built with about 1 trillion parameters and is aimed at tasks including agentic coding and cybersecurity, alongside multimodal capabilities such as image input. Several sources also say it is offered first via Mistral’s API or cloud preview, with open weights scheduled for later in the month.
Reporters describe Mistral’s release as part of a broader effort to compete with leading open-weight models, including those from China. Ars Technica and others emphasize Mistral’s claim that “Le Chonk” can challenge top closed models while remaining open-weight. Other coverage highlights Mistral’s benchmarking assertions and “sovereign AI” framing, portraying the company’s approach as regionally competitive. Technical descriptions in some reports point to a mixture-of-experts design and state that the system is trained on Mistral’s own infrastructure, including NVIDIA Grace Blackwell GPUs. Across outlets, the main differences are emphasis—performance claims, target use cases, and the timing of open-weight availability.