# 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. - **Source:** [Latest news](https://brifing.si/source/77d15218-219f-472d-bf40-31b5646d2942) - **Published:** 2026-10-29 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/e02f6604-73c8-4f9c-a38e-ffc3273696f7 ## Mistral 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. ## Coding 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. ## Efficiency 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. ## Cybersecurity 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 Unlike proprietary models where access can be rescinded by frontier labs, Le Chonk allows users to maintain full control.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. ## Defensive 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. ## Open 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. ## Infrastructure 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. ## Strategic 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. ## Sources covering this story - https://www.zdnet.com/innovation/mistral-large-4-weights-release/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._