SwarmRSI evolves scientific agent swarms for recursive self-improvement
An arXiv paper submitted on 14 Oct 2026 introduces SwarmRSI, a framework that evolves scientific agent swarms so agents improve their research capabilities through their own experimental experience. It combines Feedback-Guided Swarm Mutation, which refines hypotheses and their implementations using feedback from other agents, with Complementary Swarm Recombination, which reuses retained hypotheses and complementary program components so partial results can support new research; Recursive Swarm Evolution ties both into a feedback loop. The authors report state-of-the-art results on OpenProblems, NatureBench and MLE-Bench Lite, say accumulated research experience improves performance on held-out tasks, and note that biologists reproduced its programs for follow-up biological analyses.
It extends the agent-swarm and self-improvement idea into scientific discovery pipelines, which could reshape how automated research agents are evaluated and how they collaborate.