Reproduce “AmbiHTMLBench: A Benchmark for HTML-based Visually Rich Document parsing” 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
Boyu Niu, Yingjie Xiong, Shurui Li, Linhan Pang, Yuntong Zhang, Han Wang, Hao Wu, Kai Zuo and 3 more
The authors introduce a benchmark for reconstructing document images as HTML and evaluating both their content and visual organization. It measures text, table, and layout fidelity, alongside preference judgments. Across 1,115 pages, Gemini 3.1 Pro Preview leads the objective leaderboard with 0.8092, while GPT-5.6-Sol leads the subjective ranking; 8,128 of its 10,000 comparisons are proxy-scored, not human-labeled.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.