Currently listed through these providers:
Model details
Devstral Small 2
Devstral Small 2 is a 24-billion parameter model purpose-built for software engineering tasks that require working across large, interconnected codebases. Its design centers on tool use and multi-file editing capabilities, allowing it to explore repository structures, understand framework dependencies, and make coordinated changes across multiple files. The model supports function calling and built-in tools that enable agents to execute complex development workflows, track state across files, and retry with corrections when initial attempts fail. This architecture-level reasoning makes it well-suited for tasks like complex refactoring, bug fixing, and modernizing legacy systems, where understanding context and dependencies matters more than isolated code generation.
The model family includes the larger Devstral 2 123B alongside this smaller variant, and both versions share the same focus on agentic coding. Devstral Small 2 achieves strong performance on real-world software engineering benchmarks, with evaluations showing competitive results against leading closed models in tool-calling success rates and task completion. Its open-source availability under a permissive license means developers can run it locally for cost efficiency while maintaining the capability advantages of larger systems. The model integrates with popular coding agents and development tools, making it practical for teams looking to automate multi-step engineering tasks without relying on external API services for every operation.
Quick Info
Powered by- Provider
- Vercel AI Gateway
- Model key
- mistral/devstral-small-2
- Release date
- Dec 9, 2025
- Last updated
- May 7, 2025
- Knowledge cutoff
- 2024-10
- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $0.10
- Output token cost
- $0.30
Limits
- Output tokens
- 256,000 tokens
- Context window
- 256,000 tokens
Latest news about Devstral Small 2
No articles yet. Fetch the latest news to show it here.