Mistral Previews 1-Trillion Parameter "Le Chonk" Open-Weights AI Model

Mistral is previewing ML4, a multimodal model with 1-trillion total parameters that will be published under a custom open-weights license on October 27. The model operates on a sparse architecture, activating only 49 billion parameters during inference to maintain efficiency. Training took roughly two months using 4,000 Nvidia Grace Blackwell GPUs in Mistral’s European data centers across more than 160 languages. Mistral is explicitly targeting cybersecurity and government use cases by pitching ML4 for sovereign, zero-data-retention environments. The company recently secured a €3 billion Series D funding round at a post-money valuation above €21 billion. "ML4 is at the frontier of open weight models." — Guillaume Lample