Reproduce “EvoGroup: A Self-Evolving Harness for Group Agents in Shared Contexts” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinyi Xu, Xiaoyang Li, Yucheng Feng, Wenqing Chen, Youliang Yuan, Wei Shi, Rongxiang Weng, Zibin Zheng
EvoGroup evolves an agent’s harness—the tools, memory, prompts, and policies around its language model—using task outcomes and interaction traces to diagnose problems and revise how it retrieves and acts. On held-out EverMemBench with DeepSeek-V4-Flash, pass@1 rises from 71.07% to 89.58%, exceeding Codex and Claude Code; on GPT-5.5, its score falls slightly below both.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.