Reproduce “SWE-VERIFIER: Selective Verification for Cost-Efficient 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

Kaijian Zou, Vaibhav Chennareddy, Yunxiang Zhang, Naihao Deng, Longju Bai, Jiaxin Pei, Rada Mihalcea, Lu Wang

The method selectively verifies coding agents’ proposed actions on repository tasks. A Qwen3.5-4B model first scores an action, sampling and ranking alternatives only when its score is low. On 500 SWE-Bench Verified tasks with MiniMax-M2.7, it resolved 77.20% versus 75.47% without verification, at 1.09× normalized cost; this result is limited to the reported setup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.