Reproduce “PhySimCode: A Benchmark and Evaluation Method for Physics Video to Code Generation” 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
Souraja Kundu, Aditya Gupta, Joel Julin, Yizhou Zhao, Liuyue Xie, Laszlo A. Jeni
PhySimCode tests whether multimodal models can infer physics and parameters from simulation videos and reproduce them in Python. It provides 160,614 examples across 162 phenomena and evaluates ten models on 2,430 clips. Gemma-3-12B-IT scores 4.93/5 on law correctness, while GPT-5 Mini scores 2.47/5 on video-specific law equivalence. Claude Sonnet 4.6 recovers 7.69% of eligible parameters within tolerance in clean, synthetic scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.