NanoGPT
Devstral by MistralAI is based on Mistral Small 3.1. Debuts as the #1 open source model on SWE-bench.
Model details
Devstral Small 2505 sits inside Mistral's devstral family as a focused engineering assistant, carrying forward the architecture lineage of Mistral Small 3.1 and entering the field as the top-ranked open-source model on SWE-bench at debut. That benchmark positioning signals a deliberate emphasis on real-world software engineering tasks such as code navigation, multi-file edits, and pull-request reasoning, rather than broad general-purpose chat. The model works with text on both ends, so workflows revolve around code, documentation, and structured instructions rather than images or audio.
In practical terms, Devstral Small 2505 is well suited to agentic coding setups that need tool use and tunable sampling, with documented support for function calling and temperature control that let developers wire it into retrieval, shell, or editor pipelines. Third-party technical references describe a generous 128,000-token context window paired with an equally large output ceiling, which is helpful for repository-scale reasoning and long refactoring tasks where the model needs to hold many files in view at once. Teams looking for a code-leaning model that combines open-weight availability with a strong SWE-bench showing will find Devstral Small 2505 a natural fit for IDE assistants, autonomous debugging agents, and other developer-tooling scenarios.
NanoGPT
Devstral by MistralAI is based on Mistral Small 3.1. Debuts as the #1 open source model on SWE-bench.