reach-vb/Devstral-Small-2505

TEXT GENERATIONPricing:Input $0.7 / Cached $0.04 / Output $1.16Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 21, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Devstral-Small-2505 is a 24 billion parameter agentic LLM developed by Mistral AI and All Hands AI, fine-tuned from Mistral-Small-3.1. It is specifically designed for software engineering tasks, excelling at using tools to explore codebases and editing multiple files. The model achieves a 46.8% score on SWE-Bench Verified, positioning it as a leading open-source model for code generation and agentic capabilities, and features a 128k token context window.

Loading preview...

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, enabling it to effectively explore codebases, modify multiple files, and power sophisticated software engineering agents. It boasts a substantial 128k token context window, facilitating comprehensive code understanding and generation.

Key Capabilities

  • Agentic Coding: Optimized for autonomous software engineering tasks, making it ideal for agent-based development workflows.
  • High Performance on SWE-Bench: Achieves a verified score of 46.8% on SWE-Bench, outperforming larger models like GPT-4.1-mini and Claude 3.5 Haiku when evaluated under the OpenHands Scaffold.
  • Lightweight Deployment: Despite its capabilities, the 24B parameter model is designed to run efficiently on consumer-grade hardware, such as a single RTX 4090 or a Mac with 32GB RAM, supporting local and on-device inference.
  • Apache 2.0 License: Offers broad commercial and non-commercial use and modification rights.
  • Tekken Tokenizer: Utilizes a specialized Tekken tokenizer with a 131k vocabulary size.

Good For

  • Software Engineering Agents: Building intelligent agents that can autonomously perform coding tasks, bug fixes, and feature development.
  • Code Exploration and Modification: Tasks requiring deep understanding and manipulation of large codebases.
  • Local Development: Developers seeking a powerful yet resource-efficient model for local inference and integration into their workflows.
  • Research and Development: Experimenting with advanced agentic LLM capabilities in a software engineering context.

Finetunes
1 models