Mistral AI launched Mistral Vibe on December 9, 2025 as both a terminal coding agent and the primary delivery vehicle for the Devstral model family — purpose-built for software engineering tasks. The initial release was a minimal CLI agent; subsequent updates through early 2026 expanded it significantly, adding a cloud execution platform that allowed sessions started locally to continue running on Mistral’s infrastructure after the developer closed their laptop.
Vibe 2.0, launched in June 2026, introduced Devstral 2 as the default model. Devstral 2 scored 77.6% on SWE-bench Verified, competitive with frontier models at significantly lower inference cost — reinforcing Mistral’s positioning of specialised coding models as a more economical alternative to general-purpose frontier models for software development tasks. The release also added a VS Code extension alongside the terminal interface.
Mistral’s entry into the coding agent space with Vibe was strategically significant: as a European AI lab with Apache 2.0-licensed models, Mistral offered organisations a path to competitive coding agent capabilities without dependence on US cloud providers. The open-source release of both the Vibe CLI and the Devstral model weights gave enterprise developers a self-hostable option that combined a capable coding agent with a verifiable, auditable model.