Upgrading agentic coding capabilities with the new Devstral models
Source
mistral.ai
Date
Context
Mistral AI's July 2025 Devstral release expanded a coding-model family designed to work inside agents that solve repository issues. The original Devstral ran inside AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → frameworks that define the interface between the model and a project's tests, as its launch account explains.
The update offered Devstral Small 1.1, a 24-billion-parameter model released under Apache 2.0, and Devstral Medium, available through Mistral's API and deployable on private infrastructure. Mistral reported SWE-benchThe standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.Full definition → scores of 53.6% and 61.6%, respectively, on software-issue resolution. These are reported benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → results, and only the Small model carries the open license. The release details both options.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
SWE-bench — The standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
