Reproduce “An Empirical Study of VLM Pipelines for Long-Document QA” 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
Kenan E. Ak, Jay Mohta, Gwang Gook Lee, Yan Xu, Dimitrios Dimitriadis
The authors compare static input, retrieval, and a six-tool agent for question answering over long documents. The agent consults a document catalog and requests relevant pages or elements. With Claude Sonnet 4.5, it scores 0.625 on MMLongBench-Doc and 0.659 on LongDocURL; its advantage is not consistent across readers or benchmarks, and LongDocURL’s raw-PDF decline mostly reflects hosted API request failures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.