Devstral-Small-2505: An Agentic LLM for Software Engineering

Devstral-Small-2505 is a 24 billion parameter language model developed through a collaboration between Mistral AI and All Hands AI. Fine-tuned from Mistral-Small-3.1, this model is specifically engineered for agentic software development tasks, demonstrating strong capabilities in codebase exploration and multi-file editing.

Key Capabilities

• Agentic Coding: Optimized for software engineering agents, enabling tool use and complex code manipulation.
• High Performance on SWE-Bench: Achieves 46.8% on SWE-Bench Verified, outperforming other open-source models and even larger proprietary models like Claude 3.5 Haiku and GPT-4.1-mini on this benchmark.
• Lightweight Deployment: Its 24B parameter size allows for local deployment on consumer-grade hardware, such as a single RTX 4090 or a Mac with 32GB RAM.
• Long Context Window: Inherits a 128k token context window from its base model, Mistral-Small-3.1, though the current version specifies 32K tokens.
• Apache 2.0 License: Offers flexibility for both commercial and non-commercial use.

Good For

• Software Engineering Agents: Ideal for building autonomous agents that interact with codebases.
• Code Generation and Refactoring: Excels at tasks requiring understanding and modification of code.
• Local Development: Suitable for developers needing powerful coding assistance on local machines due to its efficient size.
• Benchmarking: A strong contender for tasks evaluated by SWE-Bench, showcasing its problem-solving abilities in real-world coding scenarios.