Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race
wired ·
French AI lab Mistral claims its latest open-weight model, Le Chonk, has surpassed leading Chinese competitors in coding and cybersecurity tasks. The trillion-parameter model was refined using increasingly difficult security challenges, allowing it to match or exceed outputs from DeepSeek and Z.ai. Pierre Stock, Mistral's VP of science, describes this achievement as a pivotal moment for Western enterprises and governments facing rising cyber threats. The company emphasizes the strategic importance of having accessible, non-proprietary models that cannot be remotely disabled by foreign entities. This release follows recent high-profile incidents where AI systems exposed vulnerabilities and raised concerns about autonomous hacking capabilities. While Mistral promotes open weights as essential for secure infrastructure, experts note the dual-use nature of such technology. Critics warn that unrestricted access allows threat actors to download and customize these powerful models for malicious purposes.
wired