French Mistral AI has unveiled its new flagship model, the Mistral Large 4, which the company unofficially calls Le Chonk. According to an independent rating by Artificial Analysis, it has become the smartest AI model created outside the US and China.
French Mistral AI has unveiled its new flagship model, the Mistral Large 4, which the company unofficially calls Le Chonk. According to an independent rating by Artificial Analysis, it has become the smartest AI model created outside the US and China.
This is reported by Artificial Analysis.
In terms of overall performance, Large 4 was approximately at the level of GPT-6 Luna and slightly behind the Chinese DeepSeek V4.1. The model showed itself especially strongly in cybersecurity. Large 4 showed itself especially strongly in cybersecurity. In one of the profile tests, the model scored 82% and was ahead of MiMo-V2.6-Pro and GPT-6 Luna
Before fully opening the model, the company decided to further test its capabilities. Now, cybersecurity experts and government officials can access the Large 4 version with less stringent security restrictions.
Large 4 can work with text, images, and documents, write code, and perform agent tasks. It is built on the Mixture of Experts architecture: the model has about a trillion parameters, but only uses a fraction of them during a single query. This allows you to avoid using all the capabilities of the model every time.
The model is now available through Mistral Studio. And on October 27, the company plans to open its download options. After that, Large 4 will be able to run on your own infrastructure, and not just use it through a cloud service.
Previously, dev.ua wrote about the previous generation Mistral Large 3, which the French startup introduced at the end of 2025 along with 9 other AI models.
