French AI startup Mistral AI has strengthened its position in the developer tools market with the release of Devstral, an open-source AI model specifically designed for software engineering tasks.
Developed in partnership with All Hands AI, creators of Open Devin, Devstral represents a significant advancement in AI-powered coding. The 24-billion parameter model achieves a 46.8% success rate on SWE-Bench Verified, a benchmark of 500 real-world GitHub issues, outperforming other open models including Google's Gemma 3 27B and DeepSeek's V3.
Unlike traditional language models focused on code completion, Devstral is optimized to function as a complete software engineering agent. "Devstral excels at using tools to explore codebases, editing multiple files and powering software engineering agents," Mistral explained in their announcement. The model can understand context across multiple files, navigate large codebases, and resolve complex issues in real-world projects.
A key advantage of Devstral is its efficiency and accessibility. The model can run on relatively modest hardware, including a single NVIDIA RTX 4090 GPU or a Mac with 32GB RAM, making it practical for local deployment and privacy-sensitive applications. This stands in contrast to many larger AI models that require substantial computing resources.
Devstral is available under the Apache 2.0 license, allowing unrestricted commercial use. Developers can download it from platforms like Hugging Face, Ollama, and Kaggle, or access it through Mistral's API at $0.1 per million input tokens and $0.3 per million output tokens.
This release marks Mistral's third product launch in May 2025, following Mistral Medium 3 and Le Chat Enterprise. The company has indicated they are already working on a larger agentic coding model to be released in the coming weeks, continuing their rapid expansion in the AI space backed by over €1.1 billion in funding.