Huawei's SwarmRSI: evolving agent swarms achieve recursive self-improvement in science
Qianru Meng · hf · 2026-10-15
Researchers from Huawei, Leiden University and University of Groningen introduce SwarmRSI, a framework for recursive self-improvement (RSI) via evolving scientific agent swarms, achieving SOTA on OpenProblems, NatureBench and MLE-Bench Lite.
Two challenges addressed:
- Poor experimental results leave it unclear whether to revise a hypothesis or its implementation;
- Overall program scores can hide components useful to other agents.
Key mechanisms:
- Feedback-Guided Swarm Mutation: refines hypotheses and implementations using experimental feedback from other agents;
- Complementary Swarm Recombination: reuses retained hypotheses and complementary program components across agents;
- Recursive Swarm Evolution integrates both into a feedback loop.
Accumulated research experience improves performance on held-out tasks, and biologists have reproduced its programs for follow-up analyses.
More from coding & agent
- t3 code demo: switching among 78 themes with the /theme command — NERDDISCO · 2026-10-15
- Dev lets Codex autonomously take over his MacBook and iPhone Air to profile and optimize his game — Dimillian · 2026-10-15
- 48 prompts, 1,000 reasoning steps: dev builds 7,342-line zero-dependency game with Gemini 4 Argon — zacharynado · 2026-10-15
- Zenzap turns group chats into control panels for AI agents with in-chat approvals — rohanpaul_ai · 2026-10-15
- React Bits 2.0 launches with 216 animated components and rebuilt design tools — shadcn · 2026-10-15
- TorchLean expands into a verified tensor programming framework with einsum and GPU execution in Lean — burny_tech · 2026-10-15