Mistral Large 4, a one-trillion-parameter open-weight model, to release weights on October 27
Mistral announced its largest model, internally called Le Chonk and officially Mistral Large 4, with one trillion parameters and 49 billion active, trained on about 3,800 Grace Blackwell GPUs, with weights released October 27.
🚨 Episode 104 of The Intelligence Journal is here. Le Chonk: A Trillion Parameters You Can Take Home 🇪🇺🐱 Mistral just announced its biggest model ever. Internally, they call it Le Chonk. Officially it's Mistral Large 4. That's one trillion parameters, 49 billion active at any moment, trained from scratch on about 3,800 Nvidia Grace Blackwell GPUs in Mistral's own European data centres. It reads text and images in more than 160 languages 🧠 On October 27, Mistral releases the weights. Anyone can download it and run it on their own machines, so their data never touches someone else's cloud 🔓 Mistral calls it the strongest open-weight model built outside China. On its own numbers, it edges GLM-5.3 and DeepSeek-V4-Pro at agentic coding and beats GPT-6 Astra on legal and finance agent tasks. Arthur Mensch, speaking in Abu Dhabi: "The narrative that Europe cannot compete is something that is not true" 🇪🇺 The twist: for the next three weeks, vetted cyber teams and state authorities get "the same model with reduced moderation and expanded cyber capabilities" ⚠️ Join us as we break down: - What a trillion-parameter model you can run on your own hardware means for banks, governments and anyone who can't send data to a US cloud. - Whether Europe has really closed the gap, or has mostly overtaken China's open models on the benchmarks Mistral picked. - Why, in the same week US labs paused tool use and signed safety pledges, Mistral is handing governments a less restricted version and then releasing the weights to everyone. 🗓️ Date: Oct 07, 2026 ⏰ Time: 3:00 PM UTC 📍 Set your reminder: https://x.com/i/spaces/1qKDzWEkEbkJV?s=20