GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning

2025-07-25

AI Summary

To address the low sampling efficiency of reinforcement learning methods in downstream adaptation of large language models (LLMs), this paper proposes GEPA—a natural-language-based reflective prompt optimization framework. GEPA diagnoses issues by analyzing system-level reasoning traces, generates improved prompts via natural language generation, and integrates genetic algorithms with Pareto frontier selection to enable multi-objective, few-shot prompt search. Its key innovation lies in the first coupling of linguistic reflection mechanisms with genetic-Pareto optimization, supporting runtime collaborative optimization across multiple LLMs. Experiments demonstrate that GEPA achieves an average 10% performance gain over GRPO across four tasks (up to +20%), with a 35× improvement in sampling efficiency; it also outperforms MIPROv2 by over 10% on two large-scale models.

Technology Category

• Search and Optimization: Learning to Search
• Natural Language Processing: Prompt Engineering / Prompting
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language can often provide a much richer learning medium for LLMs, compared with policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing one or more LLM prompts, GEPA samples system-level trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems, propose and test prompt updates, and combine complementary lessons from the Pareto frontier of its own attempts. As a result of GEPA's design, it can often turn even just a few rollouts into a large quality gain. Across four tasks, GEPA outperforms GRPO by 10% on average and by up to 20%, while using up to 35x fewer rollouts. GEPA also outperforms the leading prompt optimizer, MIPROv2, by over 10% across two LLMs, and demonstrates promising results as an inference-time search strategy for code optimization.

Problem

• Optimizing LLM prompts using natural language reflection
• Reducing rollout requirements compared to reinforcement learning
• Improving performance over existing prompt optimizers

Innovation

• Uses natural language reflection for prompt optimization
• Combines lessons from Pareto frontier attempts
• Achieves high quality gains with few rollouts