Reproduce “S-Agent 2.0: Spatial Reasoning with a General Coding Agent” 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
Rui Gong, Ruoyu Wang, Yalun Dai, Hao Li, Sidi Yang, Qihao Zhao, Hongyu Li, Zeyang Guan and 4 more
S-Agent 2.0 grounds spatial answers by combining tool measurements with model-perceived entities in a persistent scene and returning readouts that state their reference frame. On ReVSI, Qwen3.8-27B with the general coding agent scores 82.3 with the toolkit, versus 61.8 without it, a 20.5-point gain. Gains shrink with stronger backbones, and absolute-distance questions remain an exception.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.