Reproduce “EvoBrowseComp: Benchmarking Search Agents on Evolving Knowledge” 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
Yunhan Wang, Jiaan Wang, Lianzhe Huang, Xianfeng Zeng, Fandong Meng
EvoBrowseComp tests whether search agents can connect scattered, fresh web evidence rather than recall fixed answers. It generates questions from live-web facts, filters evidence, and checks reasoning structure; the English and Chinese sets each contain 400 questions and are intended for annual updates. On English questions, Claude-Opus-4.8 scored 46.2% with tools and 9.0% without, as judged by GLM-5-Chat.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.