Group-Evolving Agents wins COLM 2026 award, hits 71% on SWE-bench Verified
xwang_lk · x · 2026-10-10
Group-Evolving Agents (GEA) won the Outstanding Paper Award at the COLM 2026 Lifelong Agents Workshop. GEA treats a group of agents as the unit of evolution, sharing experience across the group to overcome the isolation of tree-structured self-evolving methods. It scores 71.0% vs 56.7% on SWE-bench Verified and 88.3% vs 68.3% on Polyglot against SOTA self-evolving baselines, matches or beats top human-designed agent frameworks, fixes framework-level bugs in 1.4 iterations on average, and transfers across coding models. First project of UCSB PhD student Zhaotian Weng (arXiv:2602.04837).
More from coding & agent
- Startup says Devin now handles 90% of dev work via Slack, displacing Codex and Claude Code — RileyRalmuto · 2026-10-10
- Charging $15k to Clone Founders With AI: The 3-Tool Local Stack That Works — aryanXmahajan · 2026-10-10
- In one year, generative UI grew from experiment to a full web protocol stack — agihouse_org · 2026-10-10
- Notes from an NYC AI dinner: agents find most inference perf wins, code review deemed unproductive — paulnovosad · 2026-10-10
- Playing SNES Classics in 4K with CRT Filters Built via Codex — pvncher · 2026-10-10
- AI Agent Gets Credit in Its Own Name: Priors' Trading Bot Now Trades on Robinhood — econoar · 2026-10-10