Reproduce “Multi-Agent Relay Thinking” 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
Xu Liu, Jungang Qiao, Guikun Chen, Xiaojuan Qi, Wenguan Wang
MACR lets multiple vision-language models continue a shared reasoning trajectory, handing it off when the active model’s local confidence falls. It standardizes confidence across models and selects a continuation through trial generations. On MMStar, MACR reaches 72.00% accuracy versus 68.13% for the strongest single-model strategy, while using fewer tokens than H-DMAD; it does not lead H-DMAD in accuracy there.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.