Mistral bets big on “vibe coding” with new autonomous software-engineering agent (arstechnica.com)

French AI startup Mistral AI has unveiled Devstral 2, a powerful 123 billion parameter open-weights coding model designed to function as part of an autonomous software engineering agent. Achieving a 72.2 percent score on the SWE-bench Verified benchmark, which tests AI systems on real GitHub issues, Devstral 2 positions Mistral among the top performers in the realm of coding models. Alongside this release, Mistral introduced Mistral Vibe, a command line interface (CLI) that enables developers to interact with the Devstral models directly within their terminals, enhancing productivity by managing file structures and executing commands autonomously. This development is significant for the AI/ML community as it underscores the trend toward creating AI tools that not only assist but also independently manage software engineering tasks. While some researchers caution that SWE-bench may oversimplify real-world coding challenges—often focusing on simpler bug fixes—it remains a crucial metric for evaluating AI capabilities. Mistral's Devstral models, including the smaller 24 billion parameter Devstral Small 2, support a generous 256,000 token context window, making them versatile tools for processing codebases of varying complexity. The licensing of both models under permissive terms further encourages innovation and collaboration within the development community.