Mistral Large 4: 1 Trillion Parameters, Open Weights This Month
3 min readMistral AI launched a public preview of Mistral Large 4 on October 6, a 1 trillion-parameter model the Paris lab has nicknamed “le Chonk.” The company says it is the strongest open-weight model ever built in the US or Europe, and it plans to release the weights by the end of October.
Why the open-weight race matters
For most of 2026 the open-weight frontier has belonged to Chinese labs. DeepSeek, Zhipu’s GLM series, Moonshot’s Kimi and Alibaba’s Qwen have set the pace for models anyone can download and run on their own hardware, while US and European labs mostly kept their best models behind an API.
Mistral has positioned itself as Europe’s answer to that gap, backed by a 3 billion euro Series D, the largest equity round ever raised by a European technology company. Mistral Large 4 is the first model funded by that round.
What Mistral shipped
According to Mistral’s announcement, Large 4 is a natively multimodal mixture-of-experts model with 1 trillion total parameters and 49 billion active per token. It was trained from scratch on 3,800 Nvidia Grace Blackwell GPUs in Mistral’s own European data centers, and the preview is served from that same infrastructure. Training data spanned more than 160 languages, including every official language of the European Union.
The preview API is live on Mistral Studio at $1.36 per million input tokens and $4.18 per million output tokens. On the Artificial Analysis coding agent harness it scores 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4, which Mistral says puts it ahead of DeepSeek V4 Pro and Qwen3.8 Max. In a blind human evaluation run by Surge AI, it ranked second of five models, behind only Claude Opus 5.
The headline capability is cybersecurity. Mistral says the model solves 93% of Cybench challenges and scores 82% on a test that asks a model to reproduce and then patch a real open-source vulnerability, the highest of any model. Several closed models score near zero on that test because they refuse the task. Until the weights drop, Mistral is red-teaming the model with security firms and state authorities who get a version with reduced moderation.
Why Mistral Large 4 matters
Mistral is making an explicit sovereignty pitch: a frontier-class model trained, hosted and governed in Europe, with weights customers can run on their own servers. For security teams, the argument is that a model which refuses to reproduce an exploit cannot help defend against it, and Mistral is betting that open weights plus self-deployment will win that market.
The model still trails the best Chinese open models on the broad Artificial Analysis Intelligence Index, where the preview scored 38, and Mistral admits the reinforcement learning run behind it is still in flight. Watch for the weight release later this month and whether the gap to GLM and Kimi closes as more compute comes online.
Mistral Large 4 is a statement that Europe intends to compete at the open-weight frontier rather than license it. Whether le Chonk can hold that position against the next DeepSeek release is the question for the rest of the year.