Devstral Small 2 is a compact agentic language model designed for software engineering workflows, excelling at tool usage, codebase exploration, and multi-file editing. With 24B parameters and FP8 instruct tuning, it delivers strong instruction following while remaining lightweight enough for local and on-device deployment. The model achieves competitive performance on SWE-bench, validating its effectiveness for real-world coding and automation tasks. It introduces vision capabilities, enabling image understanding alongside text for more versatile development workflows. Devstral Small 2 supports a 256k context window, allowing it to reason across large repositories, long diffs, and extended technical contexts. Its architecture improves generalization across diverse prompts and coding environments while leveraging advanced attention scaling techniques.
Features
- Agentic coding model optimized for software engineering automation
- 24B-parameter FP8 instruct model suitable for local and on-device deployment
- Runs on a single RTX 4090 or a Mac with 32GB RAM
- Vision capabilities for image analysis and multimodal workflows
- Large 256k context window for deep repository understanding
- Improved generalization across diverse prompts and coding environments
- Advanced attention scaling using rope-scaling and scalable softmax
- Open-source under the Apache 2.0 license for commercial and non-commercial use