Reproduce “RxnExBench: Benchmarking Vision-Language Models for Structured Chemical Reaction Extraction from Scientific Literature” 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
Jiaxing Sun, Junyuan Gao, Ruikang Zhang, Hao Zheng, Lingli Ge, Xiao An, Jiang Wu, Huazheng Zeng and 11 more
The authors introduce RxnExBench, a benchmark testing whether models can assemble reaction records from chemistry figures that combine templates, entries, and footnotes. It covers Screen Tables and Substrate Scopes, including entries containing multiple reactions. On a subset, GPT-5.5’s reaction-level F1 rises from 23.16% to 42.67% when molecular structures are supplied, but reaches only 15.29% under the stricter ten-field criterion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.