Devstral the listed price is an open-weights large language model purpose-built for agentic software engineering work, designed to explore codebases with tools, edit multiple files, and serve as the engine behind autonomous SWE agents. The released Instruct checkpoint is distributed in FP8 precision on Hugging Face under the Mistral organisation, fine-tuned for instruction following, chat, and instruction-based SWE workflows. Its 256k token context window positions it for handling substantial repositories in a single session, and the model card describes it as a step-up over prior Devstral variants, with better generalisation across diverse coding prompts and environments.

In benchmark comparisons cited by Mistral, the 123B-parameter Devstral the listed price delivers strong results on agentic coding evaluations, scoring 72.2% on SWE-Bench Verified, 61.3% on SWE-Bench Multilingual, and 32.6% on Terminal-Bench the listed price, placing it ahead of smaller open peers and competitive with much larger frontier systems on these agentic metrics. The launch coincided with Mistral's Mistral Vibe CLI, reflecting a focus on pairing the model with tooling for real coding workflows rather than standalone chat. Practically, Devstral the listed price is a good fit for teams building AI code assistants, multi-file refactoring agents, or any pipeline that needs an open, locally deployable coding model with a long context and strong tool-use orientation.