s³: Simple Scaling of Swarms via Cooperative Multi-agent RL
Published in Preprint, 2026
Can a swarm of LLM agents with no leader, no roles, and no turn order learn to organize itself? With one shared reward and fully asynchronous communication, s³ swarms learn emergent coordination, beat best-of-N, and keep working when deployed with more agents than they were trained with.
Download here
