GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning

ICLROral2026

Authors: Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Arnav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, Christopher Potts, Koushik Sen, Alex Dimakis, Ion Stoica, Dan Klein, Matei Zaharia, Omar Khattab
Affiliation: University of California, Berkeley
Venue: ICLR 2026
Track: Oral

TL;DR

GEPA uses natural language reflection to optimize prompts, outperforming GRPO and MIPROv2 while needing far fewer rollouts.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

reinforcement learning

← All ICLR 2026 Oral papers · Browse the whole archive