GEPA-Curator: Curated Learning for Self Improving Agents
arXiv cs.CL · 28 Oct 2026, 03:58 AEST
GEPA-Curator is a new framework designed to enhance prompt optimization for large language models in complex environments. It focuses on curating tasks to improve learning signals and performance outcomes.
Key facts
- Large Language Models (LLMs) are increasingly used in multi-turn agentic workloads that require sustained interaction with complex environments.
- We introduce GEPA-Curator, a framework that augments the prompt optimizer to select which agent experiences should shape prompt optimization.
- In particular, GEPA-Curator consists of two components: task curation, which filters out tasks that are likely to provide noisy or unreliable learning signals, and prior success retrieval.
- We evaluate GEPA-Curator on GoBrowse, ALFWorld, AppWorld, and $τ$-bench, using both Gemini 3.1 Pro and Claude Opus 5 as optimizer models.
- GEPA-Curator matches or exceeds the mean performance of both baselines, with gains of up to 20.7% over GEPA and 21.2% over ACE.
Read aloud
GEPA-Curator is a new framework designed to enhance prompt optimization for large language models in complex environments. It focuses on curating tasks to improve learning signals and performance outcomes. Large Language Models L L M s are increasingly used in multi-turn agentic workloads that require sustained interaction with complex environments. We introduce GEPA-Curator, a framework that augments the prompt optimizer to select which agent experiences should shape prompt optimization. In particular, GEPA-Curator consists of two components: task curation, which filters out tasks that are likely to provide noisy or unreliable learning signals, and prior success retrieval. We evaluate GEPA-Curator on GoBrowse, ALFWorld, AppWorld, and $τ$-bench, using both Gemini three point one Pro and Claude Opus five as optimizer models. GEPA-Curator matches or exceeds the mean performance of both baselines, with gains of up to twenty point seven percent over GEPA and twenty one point two percent over ACE.
Sources in this cluster
- GEPA-Curator: Curated Learning for Self Improving Agents
arXiv cs.CL · 28 Oct 2026, 03:58 AEST