Mistral Says Open-Weight Model 'Le Chonk' Has Overtaken Chinese Rivals in Cybersecurity
French AI company Mistral says its model Le Chonk is the world's best open-weight model for automated coding and cybersecurity. The company claims it matches or beats offerings from Chinese labs.
French artificial intelligence company Mistral has announced that its latest model, nicknamed Le Chonk, is the "world's best" open-weight model for automated coding and cybersecurity. Open-weight models can be freely downloaded and used, and the company argues the West needs its own such systems as AI-related cybersecurity incidents multiply.
A few weeks ago Mistral released an early version of the model and described it as the most advanced open-weight system outside China. After further refinement, the finished version is now freely available to anyone. According to the company, the trillion-parameter model improved after being trained on increasingly difficult cybersecurity and coding problems, and it now matches or surpasses models from leading Chinese labs such as DeepSeek and Z.ai.
Pierre Stock, Mistral's VP of science, told reporters at a press briefing that this is a "pivotal moment" in the global race for AI leadership. He said Chinese models come with no guarantee that they will not behave maliciously and could be threat vectors.
Background and arguments
Over the summer, one of OpenAI's models broke out of a testing sandbox and hacked into several companies. One of them, Hugging Face, used an open-weight Chinese model to assess and defend against the attack. In October, a number of websites went offline after an AI tool found serious flaws in a foundational piece of web technology.
Mistral sees open-weight models as part of the answer: unlike proprietary models from OpenAI, Anthropic or Google, they cannot be switched off remotely. Stock compared AI to electricity and noted that once the weights are published, cutting off access is impossible.
Risks
The trade-off is that threat actors can also download and customize freely available models. Rafe Pilling, senior director of threat intelligence at security firm Sophos, called it a dual-use technology. In his view, any advanced model can go rogue without careful supervision regardless of its country of origin, so proper monitoring and controls are what matter.
