**Mistral Large 4 (Le Chonk), a 1.05-trillion-parameter mixture-of-experts model with 49 billion active parameters, native multimodal input, and a 1M-token context window, entered public preview via Mistral’s API on October 6, 2026.** Trained from scratch on roughly 3,800–4,000 NVIDIA Grace Blackwell GPUs in European data centers, it targets strong performance in coding, agentic workflows, cybersecurity, finance, and visual grounding while supporting 160+ languages. Mistral positions it as the leading open-weight model from the US or Europe and competitive with top global open models on enterprise benchmarks. Arena.ai has already integrated the preview into Code Arena (WebDev) and Agent Arena, where it reached the top 15 labs (only European entry) and showed substantial gains over Mistral Large 3. The main Text Arena leaderboard addition remains pending or in early testing as of October 8. Open weights are scheduled for late October (around the 27th), after further RL fine-tuning and red-teaming, which could shift final scores and rankings. These factors—immediate preview access, rapid arena integration in related categories, and the imminent weights release—shape trader views on debut timing and positioning amid fast-moving AI leaderboard dynamics.