Reproduce “s$^3$: Simple Scaling of Swarms via Cooperative Multi-agent RL” 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
Kunal Jha, Gonçalo Faria, Shuyue Stella Li, Scott Geng, Faeze Brahman, Noah A. Smith, Yulia Tsvetkov, Max Kleiman-Weiner and 1 more
s³ trains identical agents to coordinate through voluntary messages and shared rewards, without assigned roles or turn order. Across two constructed tasks, four-agent aggregation reaches 93% success, versus 13% with individual rewards; six-agent exploration exceeds six independently trained solo agents’ any-success rate by 7.29 percentage points. Tests use small Qwen3 models and one training run per condition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.