tech
December 11, 2025
A new open-weights AI coding model is closing in on proprietary options
Devstral 2 model scores 72% on industry benchmark, nearing proprietary rivals.
TL;DR
- Mistral AI released Devstral 2, a 123B parameter open-weights coding model for autonomous software engineering agents.
- Devstral 2 scored 72.2% on SWE-bench Verified, a benchmark testing AI's ability to solve GitHub issues.
- Mistral also launched Mistral Vibe, a CLI tool enabling direct terminal interaction with Devstral models.
- Mistral Vibe can scan file structures, maintain project context, modify multiple files, and execute shell commands autonomously.
- Devstral Small 2, a 24B parameter version, was also released, scoring 68% on SWE-bench Verified and designed for local use.
- Both models feature a 256,000 token context window.
- Pricing details for Devstral 2 and Devstral Small 2 were provided, highlighting cost-efficiency.
- The term 'vibe coding,' coined by Andrej Karpathy, describes a programming style accepting AI-generated code without close review.
- Mistral aims for Devstral 2 to handle complex tasks like bug fixing and legacy system modernization at scale.
Continue reading
the original article