Mistral AI’s October 6 announcement of Mistral Large 4 (“le Chonk”), a 1-trillion-parameter multimodal MoE model with 49–52 billion active parameters, has sharply shaped trader views on the timing and source of any 4.1+ successor. The model entered public preview via Mistral Studio that day, with open weights scheduled for late October (around the 27th) under a custom license after safety testing with cybersecurity partners and governments. Trained from scratch on roughly 4,000 Grace Blackwell GPUs in European data centers, it targets enterprise workloads in coding, cyber defense, finance, and manufacturing while claiming leadership among non-Chinese open-weight systems and competitive visual grounding versus some closed frontier models. Ongoing reinforcement learning and the company’s €3 billion Series D underscore expectations of rapid iteration, though product timelines in frontier AI frequently slip.