Reproduce “Beyond Watching and Searching: Benchmarking Agents for Video-centered Deep Research Reports” 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

Jiahao Meng, Qize Yang, Yuxuan Wang, Yunfei Chu, Wensi Dai, Jin Xu, Yunhai Tong

The authors introduce a system that builds interactive research reports around a supplied video, linking timestamped moments to external sources. On 50 benchmark tasks, 18 model–harness configurations average 81.8/100 for presentation but 45.4 for content. A matched ablation gains 7.3 points, though extra research and compute were not controlled, so orchestration alone cannot be isolated.

Summary by alphaXiv.

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