Mistral AI announced the launch of Devstral 2, a 123‑billion‑parameter dense transformer, and its smaller 24‑billion‑parameter counterpart, Devstral Small 2. Both models are open‑source; Devstral 2 is released under a modified MIT license and Devstral Small 2 under Apache 2.0. The models support a 256K context window and achieve 72.2% on SWE‑bench Verified, positioning them as state‑of‑the‑art open‑weight coding models. Devstral 2 is marketed as up to seven times more cost‑efficient than Claude Sonnet on real‑world tasks and is available free via API during a trial period, after which pricing is $0.40/$2.00 per million tokens for input/output. Devstral Small 2 can be deployed locally on consumer hardware or accessed via API at $0.10/$0.30 per million tokens. Mistral also introduced the Mistral Vibe CLI, an open‑source command‑line coding assistant built on Devstral that can explore, modify, and execute changes across entire codebases. The CLI offers project‑aware context, file manipulation, version control, and command execution, and is released under Apache 2.0. It can be used in terminals or integrated into IDEs such as Zed via the Agent Communication Protocol. The models require 4 H100‑class GPUs for full deployment, while Devstral Small 2 supports single‑GPU operation on NVIDIA systems including DGX Spark and GeForce RTX. Mistral encourages community engagement through X/Twitter, Discord, and GitHub, and invites applications to join the team for open‑source research and interface development.
Introducing: Devstral 2 and Mistral Vibe CLI.
Mistral AI has released Devstral 2, a 123B‑parameter open‑source coding model, and Devstral Small 2, a 24B‑parameter variant, both licensed permissively. The models achieve 72.2% on SWE‑bench Verified and are positioned as cost‑efficient alternatives to closed models like Claude Sonnet. Alongside the models, Mistral introduced the Mistral Vibe CLI, a terminal‑based coding assistant that uses Devstral to automate code changes across entire projects. The CLI is open‑source under Apache 2.0 and offers file manipulation, version control, and multi‑file orchestration features.
Why it matters
The release demonstrates that compact, open‑source coding models can match or exceed larger closed models, lowering barriers for developers and fostering broader adoption of AI‑powered code assistance.