Mistral Releases Le Chonk Model With Advanced Coding Capabilities
Mistral Large 4 achieves top-tier coding performance while prioritizing user-controlled cybersecurity and data sovereignty.
1Mistral Launches Le Chonk
French AI lab Mistral released Mistral Large 4, nicknamed Le Chonk, in preview on October 6. This trillion-parameter model features open weights and introduces three distinct reasoning modes. These modes allow users to adjust performance based on task complexity, marking a significant step for Western-made open models in the global landscape.
2Coding Benchmark Performance
The model achieved a 41% score on the Terminal-Bench 4.0 coding benchmark, rising from 28% in the preview version. It now ranks alongside Codex GPT-6.1 Terra on the Artificial Analysis Coding Agent Index. Despite this progress, it remains one point behind GLM-5.3, which holds a composite score of 49.
3Efficiency and Token Usage
Mistral reports that Le Chonk rivals top open models like GLM-5.3 while utilizing 47% fewer tokens per task on average. This efficiency is critical as proprietary AI remains expensive to deploy. The model demonstrates that significant performance gains are achievable within just a few weeks of additional, targeted training cycles.
4Cybersecurity and Data Control
Mistral emphasizes cyber defense capabilities, offering customizable control and data sovereignty. Unlike proprietary models where access can be rescinded by frontier labs, Le Chonk allows users to maintain full control. The company collaborated with developers to assess these security features, mirroring practices seen at major labs like OpenAI and Anthropic.
5Defensive Security Benchmarking
Le Chonk performed competitively on the CyberGym benchmark, scoring just above DeepSeek V4.1 Flash. This test measures a model's ability to identify memory-safety vulnerabilities in C/C++ open-source projects and apply patches. These capabilities are designed to help enterprises defend against threat actors who exploit closed models to perform malicious cyberattacks.
6Open Versus Proprietary Models
The industry debate regarding model safety continues, with open models increasingly viewed as viable options for cyber defense. Mistral joined Nvidia’s Open Secure AI Alliance to support the democratization of security tools. This partnership argues that open models are essential for building collective defenses against the rising frequency of AI security incidents.
7Infrastructure and Training Scale
Training a competitive model required significant resources despite Mistral's smaller scale compared to pre-IPO giants. The model was trained from scratch on 4,000 Nvidia Grace Blackwell GPUs over two months. These operations were deployed entirely within Mistral’s own data centers located in Europe, ensuring localized control over the training process.
8Strategic Market Positioning
Mistral’s ability to produce a high-performing model with limited hardware highlights a shift in the AI race. While OpenAI utilized roughly 100,000 GPUs for GPT-6 Astra, Mistral achieved results with significantly fewer resources. This efficiency positions Le Chonk as a primary alternative for organizations prioritizing security, sovereignty, and cost-effective performance.
Original article: Latest news — Mistral Releases Le Chonk Model With Advanced Coding Capabilities