Group-Evolving Agents: Open-Ended Self-Improvement via Experience Sharing

TL;DR

This paper introduces Group-Evolving Agents (GEA), a new paradigm for open-ended self-improvement that enables experience sharing among agents, leading to significant performance gains and robustness in coding benchmarks.

Contribution

GEA treats groups of agents as the evolutionary unit, overcoming isolated evolution limitations and enhancing exploration, transferability, and robustness in self-evolving systems.

Findings

GEA outperforms state-of-the-art self-evolving methods on coding benchmarks.

GEA matches or exceeds top human-designed agent frameworks.

GEA more effectively converts exploratory diversity into long-term progress.

Abstract

Open-ended self-improving agents can autonomously modify their own structural designs to advance their capabilities and overcome the limits of pre-defined architectures, thus reducing reliance on human intervention. We introduce Group-Evolving Agents (GEA), a new paradigm for open-ended self-improvements, which treats a group of agents as the fundamental evolutionary unit, enabling explicit experience sharing and reuse within the group throughout evolution. Unlike existing open-ended self-evolving paradigms that adopt tree-structured evolution, GEA overcomes the limitation of inefficient utilization of exploratory diversity caused by isolated evolutionary branches. We evaluate GEA on challenging coding benchmarks, where it significantly outperforms state-of-the-art self-evolving methods (71.0% vs. 56.7% on SWE-bench Verified, 88.3% vs. 68.3% on Polyglot) and matches or exceeds top human-designed agent frameworks (71.8% and 52.0% on two benchmarks, respectively). Analysis reveals that GEA more effectively converts early-stage exploratory diversity into sustained, long-term progress, achieving stronger performance under the same number of evolved agents. Furthermore, GEA exhibits consistent transferability across different coding models and greater robustness, fixing framework-level bugs in 1.4 iterations on average, versus 5 for self-evolving methods.