Mistral AI, in partnership with All Hands AI, has unveiled the latest iterations of its developer-centric large language models under the Devstral 2507 series, consisting of Devstral Small 1.1 and Devstral Medium 2507.
Devstral Small 1.1 is an open model designed for local and embedded use in tasks such as program navigation and code search, achieving a 53.6% score on the SWE-Bench Verified benchmark.
Offering improved performance, the Devstral Medium 2507 model is accessible solely through the Mistral API or enterprise deployment agreements, achieving a 61.6% accuracy score on the same benchmark.
Both models support integration with code agent frameworks, providing flexible options for developers from experimentation to deployment in commercial settings, catering to various software engineering workflows.