Mistral unveils ‘Le Chonk’ to compete with Chinese open-weight AI models

The French company announced that users could expect “a new generation of specialized and optimized Mistral models” from here on [File] | Photo Credit: REUTERS Mistral on Tuesday (October 6, 2026) released the Mistral Large 4 AI model, also known as Le Chonk, to compete against the strongest open-source models, according to the company.The French AI startup company said that Le Chonk is a 1 trillion-parameter multimodal model with 49 billion active parameters, and described it as its “largest and most capable model to date”.By comparison, Chinese AI company Moonshot’s Kimi K3 model — launched this July — is a 2.8 trillion-parameter model.“The model we’re actually announcing today ‌is ⁠actually above the Chinese models on certain aspects, including cyber. So ⁠the narrative that Europe cannot compete is something that is not true,” CEO Artur Mensch told Reuters at a launch event.The Le Chonk model’s strengths include coding, agentic workflows, and multimodal understanding. Mistral noted that it “significantly” outperforms open-weight models developed in the US or Europe, and that it is fluent in over 160 languages.Other areas where it excels include cybersecurity, finance, law, and visual grounding.Mistral explained that its new model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs, in Mistral data centres in Europe.The company pitched the new model as a counterweight to the closed frontier models in the U.S. that have been criticised for a lack of accessibility, and the open-weight Chinese models from firms such as DeepSeek and Moonshot that have sparked security concerns in the West.Mistral further shared that ML4’s combined Coding Agent Index score of 49.8% put it ahead of Chinese rivals DeepSeek V4 Pro 0813 and Qwen3.8 Max.Le Chonk’s open weights are expected to release at the end of October, as per Mistral, and the model is currently available as a public preview.“We will release the weights by the end of the month. Until then, we are red-teaming the model in real-world settings with cybersecurity leaders, vetted partners, and state authorities, who will access the same model with reduced moderation and expanded cyber capabilities,” stated Mistral in a blog post.The French company announced that users could expect “a new generation of specialized and optimized Mistral models” from here on.“The pace of progress from here will be fast,” Mistral noted in the blog post, in what is a stark departure from the messages of U.S. AI company CEOs who have been calling for paced frontier development and forced slowdowns in AI capability. Published - October 07, 2026 12:51 pm IST