Reproduce “VisSearch-Bench: Evaluating Long Horizon Visual Search in Multimodal 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
ZhengXian Wu, Qianhao Nie, Zhuohong Chen, Junjie Gao, Yuyao Zhang, Yunyao Yu, Hangrui Xu, Cheng Cheng and 5 more
VisSearch-Bench tests whether multimodal agents keep visual evidence linked to the right objects and facts during open-web search. Its 204 tasks score complete answers with Accuracy and recovered answer components with Process Score. ReAct with Kimi-K3 achieves the highest Accuracy, 26.47%; GPT-5.5 + ReAct has the highest Process Score, 49.52%. These metrics distinguish complete-task success from partial recovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.