Mistral queues 'Le Chonk' open-weight 1T MoE for late October

Mistral AI has staged a Hugging Face landing page for Mistral-Large-4, the model it has nicknamed 'Le Chonk,' with a countdown toward an open-weights release scheduled for the end of October. The page itself is sparse: the model ID 'mistralai/Mistral-Large-4-1T-A52B,' a 'Natively multimodal' tag, 742 users set to be notified, and an 'Upcoming release' banner.

The specs behind the countdown already sit in Mistral's preview announcement from October 6. It is a mixture-of-experts system with 1.05 trillion total parameters and 49 billion active per token, carrying a 1.6-billion-parameter vision encoder and a one-million-token context, with support across 160+ languages. According to Slashdot's write-up of Mistral's launch post, the model was trained on 4,000 Nvidia Grace Blackwell GPUs over two months and deployed inside European data centers.

Mistral's own framing is where the hedge lives. The company calls Le Chonk 'the strongest open-weight model developed outside China by a wide margin,' a qualifier that leaves Chinese open-weight systems above the bar, and it concedes the model 'still lags behind the frontier in areas such as coding.' Guillaume Lample, Mistral's co-founder and chief scientist, pitched the security angle: 'The cyber defense capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models.'

Three of the researchers in our Who's Who directory pushed the Hugging Face listing out over the past day. Preview access went live for developers, cybersecurity teams and governments on October 6; the weights drop is slated for later this month.