gepa-pareto-evolutionary-prompt-optimizer

IN premise — summaries/2026/08/24/wiki-In-context_learning_natural_language_processing-chunk-1.md

Created 2026-08-24T17:11:12+00:00

GEPA (Agrawal 2025) combines LM-based trace analysis with Pareto-based evolutionary search over candidate prompt systems, reporting ~10% gains over GRPO and MIPROv2 with up to 35× fewer rollouts.

Summary

GEPA is a prompt-optimization method that uses language models to analyze where prompts fail and then evolves improved prompts through a multi-objective search process. It matters because it achieves a roughly 10% quality improvement over strong baselines like GRPO and MIPROv2 while needing up to 35 times fewer expensive model rollouts, making high-quality prompt tuning dramatically cheaper in compute.