Reproduce “Terminal-style Mathematical Modeling Bench: The First Agentic and Verifiable Benchmark for Mathematical Modeling” 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
Xiaojun Wu, Cehao Yang, Xuhui Jiang, Chengjin Xu, Jia Li, Jian Guo
TB-MathModel turns complete contest-style modeling problems into terminal tasks with structured numerical answers scored against Outstanding-award papers and checked for formalized feasibility constraints. Agents submit JSON artifacts; the benchmark combines reference-panel closeness with a programmatic gate and Lean proofs. In a dragon-dance audit, authors corrected an infeasible reference from 1.406276 to 1.246266 m/s, but checks do not establish physical truth.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.