Devstral launches AI-powered software development platform

Mistral AI and All Hands AI have released Devstral, a groundbreaking open-source AI model specifically designed for software engineering that outperforms existing options for coding assistance. This new lightweight yet powerful large language model (LLM) achieves significantly better results on real-world programming tasks than both open and some closed-source alternatives, while being accessible enough to run on consumer hardware. The Apache 2.0 license makes it freely available for both individual developers and enterprises needing secure, compliant AI coding assistance.

The big picture: Devstral represents a significant advancement in AI-powered software development by tackling real-world coding challenges rather than just simpler, isolated programming tasks.

• The model achieves a 46.8% success rate on SWE-Bench Verified, a benchmark of 500 actual GitHub issues, outperforming previous open-source state-of-the-art models by more than 6 percentage points.
• When compared to closed-source alternatives, Devstral demonstrates impressive results, surpassing GPT-4.1-mini by over 20% on the same benchmark.

Why this matters: Unlike typical LLMs that excel at atomic coding tasks but struggle with complex software engineering problems, Devstral is specifically designed to understand and navigate large codebases.

• The model is trained to identify relationships between disparate components and detect subtle bugs in complex functions, mirroring how professional developers approach real-world programming challenges.
• Devstral runs over code agent scaffolds like OpenHands or SWE-Agent, which provide the interface between the model and test cases, enabling it to effectively interact with code repositories.

Key technical performance: Under the same test scaffold (OpenHands), Devstral outperforms much larger models including Deepseek-V3-0324 (671B) and Qwen3 232B-A22B.

Practical applications: Devstral offers versatility across different deployment scenarios from individual developers to enterprise environments.

• The model is lightweight enough to run on a single RTX 4090 GPU or a Mac with 32GB RAM, making it suitable for local, on-device development.
• For enterprises with privacy-sensitive repositories and strict security requirements, Devstral provides a capable solution that can be deployed internally.
• The model integrates well with agentic coding IDEs, plugins, and environments, offering developers a powerful option in their AI toolkit.

Availability: Mistral AI has made Devstral widely accessible across multiple platforms starting today.

• The model is available as a free download on HuggingFace, Ollama, Kaggle, Unsloth, and LM Studio under the Apache 2.0 license.
• Developers can also access Devstral through Mistral’s API as “devstral-small-2505” at the same pricing as Mistral Small 3.1: $0.1/M input tokens and $0.3/M output tokens.
• Enterprise customers seeking customization options like fine-tuning on private codebases or model distillation can contact Mistral’s applied AI team directly.

Where we go from here: Mistral AI describes Devstral as a “research preview” and is already developing a larger agentic coding model for release in the coming weeks.