Mistral AI Devstral Release

Mistral AI and All Hands AI have introduced Devstral, an agentic large language model (LLM) specifically designed for software engineering tasks. Devstral is released under the Apache 2.0 license and is optimized to solve real-world GitHub issues by integrating with code agent scaffolds.

High Performance on SWE-Bench Verified

Devstral outperforms prior open-source state-of-the-art (SoTA) models on the SWE-Bench Verified benchmark, a dataset consisting of 500 manually screened real-world GitHub issues.

Key performance metrics include:

  • Score: Devstral achieves a 46.8% score on SWE-Bench Verified, exceeding previous open-source SoTA models by more than 6 percentage points.
  • Comparison to Large Models: When evaluated using the OpenHands test scaffold, Devstral outperforms significantly larger models, including Deepseek-V3-0324 (671B) and Qwen3 232B-A22B.
  • Comparison to Closed Models: Devstral surpasses several closed-source alternatives, including GPT-4.1-mini by over 20%.

Agentic Capabilities for Software Engineering

Unlike typical LLMs that excel at atomic coding tasks like standalone function writing or code completion, Devstral is engineered to handle complex software engineering problems. This includes:

  • Contextualization: The ability to understand code within the context of a large codebase.
  • Relationship Mapping: Identifying connections between disparate components of a system.
  • Bug Identification: Finding subtle bugs within intricate functions.

To achieve these capabilities, Devstral is trained to solve real GitHub issues and operates via agent scaffolds such as OpenHands or SWE-Agent, which manage the interface between the model and the test cases.

Deployment Versatility and Hardware Requirements

Devstral is designed for high efficiency, allowing it to be deployed across various environments:

  • Local Deployment: The model is light enough to run on a single RTX 4090 GPU or a Mac with 32GB of RAM, making it suitable for on-device use.
  • Enterprise Use: Due to its performance and local deployment capabilities, it is a viable option for agentic coding on privacy-sensitive repositories subject to strict security and compliance requirements.
  • IDE Integration: Devstral can be integrated into agentic coding IDEs, plugins, or environments as a model selector option.

Availability and Pricing

Devstral is available through multiple channels for community and enterprise use:

  • Open Source: The model is released for free under the Apache 2.0 license and can be downloaded from HuggingFace, Ollama, Kaggle, Unsloth, and LM Studio.
  • API Access: The model is available via Mistral AI's API as devstral-small-2505. Pricing is set at $0.1 per million input tokens and $0.3 per million output tokens, matching the pricing of Mistral Small 3.1.
  • Enterprise Customization: Mistral AI offers fine-tuning on private codebases, continued pre-training, and distillation services through their applied AI team.

Future Outlook

Devstral is currently a research preview. Mistral AI has announced that a larger agentic coding model is under development and is expected to be available in the coming weeks.

Sources

Related

  • Dispatch
    Mistral AI Devstral Models Release

    Mistral AI has released Devstral Medium and Devstral Small 1.1, providing state-of-the-art agentic coding capabilities with a focus on generalization across different prompts and agentic scaffolds.

  • Dispatch
    Mistral AI Codestral 25.08 and Enterprise Coding Stack Release

    Mistral AI has released Codestral 25.08 and a comprehensive enterprise coding stack featuring integrated completion, semantic search, and agentic workflows with flexible deployment options.

  • Dispatch
    Mistral AI Devstral 2 and Mistral Vibe CLI Release

    Mistral AI has released Devstral 2 (123B) and Devstral Small 2 (24B), open-weight coding models that achieve state-of-the-art performance on SWE-bench Verified with high cost-efficiency, alongside the open-source Mistral Vibe CLI for autonomous code automation.

  • Dispatch
    Mistral AI Leanstral Release

    Mistral AI has released Leanstral, an open-source code agent with 6B active parameters designed for Lean 4 to enable formally verified code and mathematical proofs.

  • Dispatch
    Mistral Small 3 Release Notes

    Mistral AI has released Mistral Small 3, a latency-optimized 24B-parameter model under the Apache 2.0 license that rivals the performance of models three times its size.