Devstral is an agentic language model for software engineering tasks built through a collaboration between Mistral AI and All Hands AI. Devstral excels at using tools to explore codebases, editing multiple files, and powering software engineering agents. The model achieves the number one performance on SWE-bench among open source models. Devstral is finetuned from Mistral-Small-3.1, giving it a long context window of up to 128k tokens. With 24 billion parameters, it is compact enough to run on a single RTX 4090 or a Mac with 32GB of RAM, making it suitable for local deployment and on-device use. The model uses an Apache 2.0 license and is built with a Tekken tokenizer featuring a 131k vocabulary size.