Reproduce “Scaling Multi-Agent Intelligence with Substrate-Centric Coordination” 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
Zhenting Qi, Zhuokai Zhao, Rulin Shao, Kerem Oktar, Zhiyuan Yao, Himabindu Lakkaraju, Wen-tau Yih, Ansong Ni and 2 more
ScaleMAS coordinates asynchronous agents through a shared, searchable record of linked findings. Agents publish work and retrieve a bounded set of relevant messages, ranked by textual relevance and later citation reuse. On BrowseComp-Plus, eight agents using Qwen3.5-9B scored 60.0%, versus 26.0% for 16 DeLM agents at similar token use; this measures benchmark Pass@N and inference tokens, not elapsed time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.