Mistral opens public preview of 1-trillion-parameter Le Chonk, weights promised by end of October
On October 6, Mistral opened a public preview of Mistral Large 4 (Le Chonk), a 1T-total / 49B-active natively multimodal MoE priced at $1.36/$4.18 per million tokens; weights are scheduled for Hugging Face release at the end of October.
On October 6, French AI company Mistral opened a public preview of Mistral Large 4, internally codenamed Le Chonk (ML4). It is the largest open-weight model the company has shipped: 1 trillion total parameters, 49 billion active per token, with a natively multimodal design. Weights are scheduled to be released on Hugging Face and other model repositories by the end of October.
Users can try the preview today through the Mistral Studio API endpoint. In parallel, Mistral is running red-team testing with leading cybersecurity firms, vetted partners and government regulators in real business workflows, with a less-restricted test build that loosens content safety filters and strengthens cyber offense-and-defense capabilities. Mistral says ML4 is state-of-the-art among open models across cybersecurity, finance and law enterprise workloads.
Training took place in Mistral's own European data centers on roughly 3,800 NVIDIA Grace Blackwell GPUs, with a corpus spanning over 160 natural languages and all official EU languages. Mistral emphasizes that the European deployment is fully end-to-end operated by Mistral under EU law and does not rely on any third-party digital service provider.
On the Artificial Analysis Intelligence Index, ML4 scores 38 points, compared with 9 for Mistral Large 3, 14 for Medium 3.5 and 58 for Claude Opus 5.5. On the new Artificial Analysis Cyber Index ML4 lands in the global top five and leads all non-Chinese open-weight models. On a test asking a model to reproduce and patch a real vulnerability in open-source software, ML4 scores 82 percent, while Claude Opus 5.5 and GPT-6 Astra score near zero due to refusal mechanisms. ML4 also passes 93.3 percent of prompt-injection attacks on the Lakera B3 AI Security Benchmark.
During the preview the API is priced at $0.68 per million input tokens, $2.09 per million output tokens and $0.07 per million cached input tokens; documentation also lists consumer rates at $1.36 input and $4.18 output per million tokens. Mistral notes that Le Chonk is only the first in a series of new models funded by its 3 billion euro Series D closed last month, and reinforcement learning is still running, so the model is expected to keep improving.