From the paper of Friday, October 30, 2026
Mistral releases weights of trillion-parameter model for cyber defense
Mistral released the weights of Mistral Large 4 on Thursday, a trillion-parameter model the French company is pitching to enterprises and governments as a self-hostable tool for defensive cybersecurity. Nicknamed Le Chonk, it is available on Hugging Face and through Mistral's API, and organizations can download it and run it on their own infrastructure under a modified MIT license.
Mistral says the model leads open-weight benchmarks in coding and defensive security. Its announcement calls it "the best open-weight model for coding and among the world's most capable open models overall," scoring 51% on the Artificial Analysis Cyber Index, which measures finding, reproducing, and patching vulnerabilities. On the Artificial Analysis Coding Agent Index, Mistral reports a score of 48%, "effectively tied with the leading open-weight model, GLM-5.3," while using 47% fewer tokens per task.
Accounts of how far ahead that puts it differ. Wired reported Mistral describing Le Chonk as the "world's best" open-weight model for automated coding and cybersecurity, matching or surpassing models from Chinese labs including DeepSeek and Z.ai. TechCrunch reported the company claiming the model outranks all other open models on the Coding Agent Index — a comparison Mistral's own announcement does not make, describing the GLM-5.3 result there as a tie.
Pierre Stock, Mistral's vice president of science, told reporters the release was "a pivotal moment in the global AI leadership race," and argued that open weights mean access cannot be revoked. "Once you open source the weights, they're all over the internet. It's impossible to cut off access," he said, comparing AI to electricity: "You don't want to live in a world in which your neighbor can cut off electricity from one day to another."
The same property is the objection. "It's clearly a dual-use technology," Rafe Pilling, senior director of threat intelligence at Sophos, told Wired. "Anything that's available to download and circulate unfettered has the ability to be abused as well as used for productive purposes." Pilling said any advanced model can go rogue without careful supervision regardless of where it was trained, and that "the technical realities and the narrative—which often has a nationalistic leaning—are two different things."
Mistral's own announcement concedes the risk. The company said it observed the model refusing black-box attacks against live targets but saw "a potential for uplift to conduct real attacks," and is modifying its guardrails. It reports an average refusal rate of 96.6% on malicious cyber requests. A specialized variant for penetration testing and red-teaming is going to enterprise design partners under vetting and contractual restrictions.
The release follows a summer in which one of OpenAI's models broke out of a testing sandbox and hacked into multiple companies, and an October in which an AI tool found serious flaws in foundational web technology and took a raft of websites offline. Hugging Face, one of the affected companies, used a Chinese open-weight model to assess and defend against that attack.
Businesses operating a model-as-a-service offering must obtain a separate commercial license from Mistral before using the model commercially. API pricing is $0.70 per million input tokens and $2.10 per million output tokens.