Mistral AI, in collaboration with All Hands AI, has released a brand-new open-source model called Devstral, designed specifically to tackle complex software engineering tasks. Unlike many general-purpose AI models that do well with simple functions or code snippets, Devstral is built to handle real-world coding challenges like understanding large codebases, fixing bugs, and resolving actual GitHub issues.

And it's not just theory Devstral has already shown strong results. On the SWE-Bench Verified benchmark (which uses 500 real GitHub issues to test coding ability), Devstral scored 46.8%, beating all other open-source models by over 6 percentage points. It even outperformed some massive closed-source models, including GPT-4.1-mini and DeepSeek-V3.

What Makes Devstral Special?

Built for Real Coding Tasks: Devstral works with tools like OpenHands or SWE-Agent that simulate developer workflows, giving it a more practical edge.

Lightweight Yet Powerful: You can run Devstral locally on a machine with an RTX 4090 GPU or a Mac with 32GB RAM.

Versatile Use Cases: It’s ideal for local development, enterprise environments with strict data rules, or integrating into smart coding IDEs and plugins.

Open-Source Friendly: Released under the Apache 2.0 license, it's free for anyone to use, customize, or deploy.

Try It Out

Devstral is now available for download on platforms like Hugging Face, Ollama, Kaggle, Unsloth, and LM Studio. For API access, you can find it under the name devstral-small-2505, priced the same as Mistral Small 3.1.

Enterprise Ready

For companies that want to fine-tune Devstral on their own private codebases or need advanced customization, the Mistral team is offering support through its applied AI team.

Check out the full documentation or tutorial video to start experimenting with Devstral today.