French AI startup Mistral has introduced a new open-source model tailored for software development: Devstral. Released under the permissive Apache 2.0 license, Devstral is built to supercharge coding workflows, offering both on-device flexibility and commercial readiness.
Unlike its predecessor Codestral—which had restrictive licensing that barred commercial use—Devstral is ready to go to work in real-world development environments. Mistral developed it in partnership with All Hands AI, and early benchmark results are promising: Devstral reportedly outperforms leading open models like Google’s Gemma 3 27B and DeepSeek’s V3 on SWE-Bench Verified, a test suite measuring real coding capabilities.
At just 24 billion parameters, Devstral is relatively lightweight. Yet it’s powerful enough to handle advanced tasks like codebase navigation, multi-file editing, and integration with agentic systems such as OpenHands and SWE-Agent. Despite its capabilities, it runs efficiently on consumer hardware—an Nvidia RTX 4090 or a Mac with 32GB RAM is enough, making it ideal for developers seeking low-latency local deployment.
This move comes at a time when demand for AI coding tools is booming. A 2024 survey found that more than 76% of developers are already using or planning to use AI in their software development process. Although AI-generated code still struggles with issues like logic errors and security vulnerabilities, many developers are willing to accept those trade-offs for the productivity gains AI can offer.
Devstral’s release signals Mistral’s deeper push into developer tools. Earlier, the company launched Codestral but limited its usability. Now, with Devstral freely available on platforms like Hugging Face and via Mistral’s own API, the company is widening access. For developers interested in API usage, pricing starts at $0.10 per million input tokens and $0.30 per million output tokens.
In the background, Mistral is already working on an even more advanced agentic coding model. While Devstral is being framed as a “research preview,” the next iteration is expected to bring larger capabilities tailored for full-fledged autonomous coding agents.
This marks Mistral’s third product rollout in a matter of weeks. The company also launched Mistral Medium 3, a compact yet capable general-purpose model, and Le Chat Enterprise, a business-focused AI assistant that integrates with Gmail, SharePoint, and Google Drive. With over €1.1 billion in funding from backers like General Catalyst, and clients including AXA, BNP Paribas, and Mirakl, Mistral is fast becoming a major force in Europe’s AI scene.
As the race to build coding-optimized AI heats up, Devstral’s open and commercially friendly license gives it a strong advantage. Developers now have another compelling option—one designed to run locally, deploy flexibly, and support real-world software engineering tasks without compromise.
