Mistral AI released Devstral 2, a 123 billion parameter open-weights coding model that scores 72.2% on SWE-bench Verified, approaching proprietary model performance. The release includes Mistral Vibe, a CLI tool for autonomous software engineering similar to Claude Code and OpenAI Codex. A smaller 24 billion parameter version (Devstral Small 2) achieves 68% on the same benchmark and can run locally on consumer hardware. Both models support 256,000 token context windows and are released under permissive licenses (modified MIT and Apache 2.0).

2m read timeFrom arstechnica.com
Post cover image
Table of contents
Ars VideoWhat Happens to the Developers When AI Can Code? | Ars Frontiers

Sort: