people · May 20, 2026

SWE-agent Baselines Used in ContraFix Agentic Vulnerability Repair Paper

The ContraFix paper on arXiv three days ago included SWE-agent as a baseline for agentic vulnerability repair. SWE-agent with Claude-3.7-Sonnet achieved a 31.5% resolution rate and 1.29 average score on vulnerability tasks. The evaluation compared it against Aider, OpenHands, and GPT-5 variants across 28 vulnerability types. SWE-agent resolved 77 out of 224 cases in one reported metric set.