Mistral AI launched Large 4, nicknamed Le Chonk, a 1-trillion-parameter sparse MoE model with 49 billion active parameters. Trained in Europe on under 4,000 GPUs, it targets cybersecurity, finance and visual tasks while promising open weights on October 27. Early benchmarks show leadership among non-Chinese open models but still trail top closed systems. The move sharpens Europe's bid for AI sovereignty.