Open-source Refact.ai achieves #1 on SWE-bench Lite with a 59.7% score.
Our approach: fully autonomous Agent, no manual intervention needed.
How we did this:
• Prompt strategy: https://github.com/smallcloudai/refact/blob/swe-boosted-prom...
• Claude 3.7 Sonnet as orchestrator
• deep_analysis() tool (powered by o4-mini) for reasoning
• Tool suite for repository exploration, code modification, and testing. Used dynamically based on task needs
• One correct solution through iteration!
Autonomy = our core strength.
Refact.ai Agent completes the entire dev workflow independently: plans, executes, tests, self-corrects, and delivers a production-ready result. For each task, it made one multi-step run to generate a single correct solution, creating custom strategies rather than following rigid scripts.
How we did this:
• Prompt strategy: https://github.com/smallcloudai/refact/blob/swe-boosted-prom... • Claude 3.7 Sonnet as orchestrator • deep_analysis() tool (powered by o4-mini) for reasoning • Tool suite for repository exploration, code modification, and testing. Used dynamically based on task needs • One correct solution through iteration!
Autonomy = our core strength.
Refact.ai Agent completes the entire dev workflow independently: plans, executes, tests, self-corrects, and delivers a production-ready result. For each task, it made one multi-step run to generate a single correct solution, creating custom strategies rather than following rigid scripts.
You can read tech details on our SWE-bench approach: https://refact.ai/blog/2025/sota-on-swe-bench-lite-open-sour...
Your questions are welcome! Also, welcome to try Refact.ai Agent in VS Code and Jet Brains: https://linktr.ee/refactai