Join waitlist
Back to products
Devstral 2

Devstral 2

SOTA open-source agentic coding models and CLI agent

Website mistral.ai
Overview

What it is

- We’re committed to empower the AI community with open technology. Our open models sets the bar for efficiency, and are available for free, with fully permissive license. - Our optimized commercial models are designed for performance and are available via our flexible deployment options.

Intent

I need it when

Automate software engineering tasks and autonomous code generation

Devstral 2 is an open-weights agentic coding model designed specifically for autonomous software engineering. It powers Vibe CLI and IDE integrations to generate code, refactor legacy systems, and handle multi-step coding workflows without manual intervention.

Reduce software development time through AI-assisted code completion and generation

Devstral 2 provides architecture-aware reasoning, legacy code translation, and automated code reviews. It enables developers to ship faster by handling routine coding tasks, test authoring, and documentation generation autonomously.

Integrate a specialized coding model into development tools and CI/CD pipelines

Devstral 2 is available as an API endpoint (devstral-medium-latest) with clear token pricing, enabling developers to embed it into VS Code, JetBrains IDEs, terminal environments, and automated testing/review systems for continuous code generation and analysis.

Deploy a cost-effective open-source coding model for enterprise software teams

Devstral 2 is released as open-weights under a modified MIT license, allowing commercial deployment with a Mistral license. At $0.4/$2 per million tokens (input/output), it offers a lower-cost alternative to proprietary coding models while maintaining agentic capabilities.

Drop

Not a fit when

  • User needs a general-purpose conversational AI assistant; Devstral 2 is specialized for autonomous coding and software engineering tasks only
  • User requires real-time interaction through a web or mobile UI; Devstral 2 is an API-based model designed for programmatic integration
  • User needs document understanding or OCR capabilities; Devstral 2 is optimized for code generation and agentic coding workflows
  • User operates on extremely tight budget with minimal token usage; per-token pricing may accumulate quickly for large-scale deployments
  • User requires a lightweight model for edge deployment; Devstral 2 is a 128B+ parameter model requiring substantial compute resources
  • User needs multilingual or multimodal capabilities beyond code; Devstral 2 is text-to-text and code-focused without vision or audio support
Commercials

Pricing

Pay-per-token API pricing; input tokens charged at $0.4 per million, output tokens at $2 per million View pricing