ZDNET’s key takeaways
- As planned, Mistral has released the weights for its ML4 model.
- ML4 sets a new bar for western open models.
- Open models are increasingly competitive with proprietary models.
French AI lab Mistral previewed its latest model, Mistral Large 4 (ML4), nicknamed “Le Chonk” for its trillion-parameter size, on Oct. 6.
Also: Open Weights vs. Closed: An AI Civil War Is
New coding performance ranking
ML4’s score on the agent coding benchmark Terminal-Bench 4.0 jumped to 41% from its preview version from 28% to 41%, beating Opencode’s GLM-5.3 by a small but important point. In the overall index, however, which shows a model’s composite score across three relevant benchmarks, ML4 is a point behind GLM-5.3, with 48 to GLM-5.3’s 49. Codex GPT-6.1 Terra also has a 48 overall.
As a reminder, benchmark scores themselves should be taken with a grain of salt, especially considering how many models cheat.
Nevertheless, this improvement in performance is meaningful for just a few weeks of extra training. As proprietary AI remains expensive to use, economical models that do not sacrifice performance are striking. ML4 “rivals top open models like GLM-5.3 in the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task on average,” Mistral said.
Cybersecurity Prowess
When the Mistral ML4 was released in preview earlier this month, it announced how the model prioritized cyber defence capabilities, customizable advertising control and data sovereignty. The company has spent the last few weeks working with developers and leaders to assess the model’s cybersecurity capabilities, a practice that is becoming common for proprietary American labs such as OpenAI, Google, and Anthropic as concerns about model security mounts.
Similar to Anthropic’s Project Glasswing and OpenAI’s rollout of Astra, Mistral’s initial test partners have access to a less protected version of the ML4 with “enhanced cybersecurity capabilities.”
At least by some measures, ML4’s promises seem to be coming true: Le Chonk just topped DeepSeek V4.1 Flash on the CyberGym benchmark for defensive cybersecurity, which measures how well models “find real memory security vulnerabilities in C/C++ open-source projects, reproduce them with proof-of-concept projects without breaking proof-of-project.”
Recent AI security incidents have somewhat pitted open and proprietary models against each other. Initially considered less secure due to their malleability, open models are seen as a new option for cyberdefense after proprietary models from Anthropic and OpenAI proved just as risky.
Mistral said ML4 is built for security that remains under user control — unlike proprietary models to which border labs can technically revoke access at any time.
“The cyber defense capabilities allow enterprises and governments to defend themselves against threat actors who jailbreak closed models to carry out cyber attacks,” Mistral co-founder Guillaume Lample said during a briefing for the October 6 preview.
After the Hugging Face breach, Mistral was one of many companies that signed Nvidia’s Open Secure AI Alliance, a cross-industry partnership that argues that open models are crucial to democratizing the defense against increasingly common AI security incidents.
Also: This new ChatGPT scam tricks you into installing malware
By the logic of Nvidia, and its alliance that aims to democratize AI security tools, the race is between Mistral and other open models to achieve state-of-the-art security capabilities.
More for less computing
Training a competitive model in a computer shortage is no small task for a trimmer lab like Mistral, which does not have the same resources as a pre-IPO giant like Anthropic.
Also: This Claude Enterprise feature lets you track AI ROI
“ML4 was trained from scratch on 4,000 Nvidia Grace Blackwell GPUs over two months, deployed in Mistral’s own data centers in Europe,” the company said. For context, Nvidia CEO Jensen Huang told X that OpenAI trained GPT-6 Astra on about 100,000 GPUs.
See full bio