Reproduce “TCSAlgBench: Benchmarking Automated Proving for Research-Level Theoretical Computer Science” 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
Chutong Yang, Xiyuan Zhang, Yu Huang, Boran Han, Soonho Kong, Shuai Zhang, Vihang Prakash Patil, Zhen Han and 2 more
TCSAlgBench tests whether language models can prove algorithmic results from research-level theoretical computer science papers. Its expert-designed pipeline restores needed context while withholding source-paper lemmas and, when relevant, named algorithms. GPT-5.6 Sol max achieved 23.6% five-run coverage; a separate GPT-5.5 xhigh planning study reached 25.4% with higher token use, on this proof task—not end-to-end research.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.