Devstral was developed by Mistral in collaboration with AI company All Hands AI. It is openly available under the Apache 2.0 license, which means it can be used commercially without restrictions. Mistral claims that Devstral outperforms other open-source models, such as Google’s Gemma 3 27B and the V3 from Chinese lab DeepSeek, in the SWE-Bench Verified test, which measures coding skills.
Devstral was developed by Mistral in collaboration with AI company All Hands AI. It is openly available under the Apache 2.0 license, which means it can be used commercially without restrictions. Mistral claims that Devstral outperforms other open-source models, such as Google’s Gemma 3 27B and the V3 from Chinese lab DeepSeek, in the SWE-Bench Verified test, which measures coding skills.
“Devstral makes great use of tools for exploring codebases, editing multiple files, and the power of software engineering agents. It runs on code agents like OpenHands or SWE-Agent, which define the interface between the model and test cases. Devstral is lightweight enough to run on a single [Nvidia] RTX 4090 graphics card or a Mac with 32GB of RAM, making it an ideal choice for local deployment and on-device use,” Mistral told TechCrunch.
Devstral can be downloaded from AI development platforms including Hugging Face, and can also be used via the Mistral API. The price is $0.1 per million input tokens and $0.3 per million output tokens.
Devstral is not a small model per se, but it is somewhat smaller in size, with 24 billion parameters.
