Reproduce “SWE-Duel: Towards a Self-Scaling Arena for Coding Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wei Minn, Zhensu Sun, Lwin Khin Shar, David Lo
SWE-Duel generates coding challenges: one agent adds a feature, hides a defect, and writes tests; another repairs it without knowing a defect exists, while retaining the feature. A self-repair gate filters tasks the author cannot solve. With one fixed reviewer, GLM-5.3 tasks defeated it on 42.9%, versus 4.0% for pooled GLM-5.1 and GLM-5.2 tasks, preliminary evidence limited to one model family.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.