群体演化智能体：通过经验共享实现开放式自我进化

摘要

开放式自我改进智能体能够自主修改其自身结构设计，以提升能力并突破预定义架构的限制，从而减少对人类干预的依赖。我们提出群体进化智能体（GEA）这一开放式自我改进新范式，将智能体群体作为基本进化单元，实现进化过程中群体内显性的经验共享与复用。与现有采用树状进化结构的开放式自我进化范式不同，GEA克服了因进化分支孤立而导致的探索多样性利用低效的局限。我们在具有挑战性的编程基准测试中评估GEA，其表现显著优于最先进的自我进化方法（在SWE-bench Verified上达到71.0%对比56.7%，在Polyglot上达到88.3%对比68.3%），并与顶尖人工设计的智能体框架持平或更优（在两个基准测试中分别达到71.8%和52.0%）。分析表明，GEA能更有效地将早期探索多样性转化为持续的长期进步，在相同进化智能体数量下实现更强性能。此外，GEA在不同编程模型间展现出稳定的可迁移性及更强鲁棒性，平均仅需1.4次迭代即可修复框架级错误，而自我进化方法需要5次迭代。

English

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.

群体演化智能体：通过经验共享实现开放式自我进化

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

摘要

Support