mlsanityData to 7 Oct 2026

GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning

arXiv 2507.19457 ↗First posted 25 Jul 2025cs.CL cs.AI cs.LGcited by 449

On the map: LLM reasoning → Reinforcement learning for reasoning

Abstract (arXiv)

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 often provides a much richer learning medium for LLMs, compared to 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 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 six tasks, GEPA outperforms GRPO by 6% 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% (e.g., +12% accuracy on AIME-2025), and demonstrates promising results as an inference-time search strategy for code optimization. We release our code at https://github.com/gepa-ai/gepa .

Key facts. Cited by 449 papers in the map (data through 7 Oct 2026). Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning. First posted on arXiv 25 Jul 2025.

Builds on

0 of 0 references.

No references here yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 58; 2026: 391.

449 papers.

Sort:

By area:

  1. 2025Agentic Context Engineering: Evolving Contexts for Self-Improving Language Modelscited by 349
  2. 2026Meta-Harness: End-to-End Optimization of Model Harnessescited by 248
  3. 2026SkillOpt: Executive Strategy for Self-Evolving Agent Skillscited by 122
  4. 2026Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnessescited by 119
  5. 2025The Attacker Moves Second: Stronger Adaptive Attacks Bypass Defenses Against Llm Jailbreaks and Prompt Injectionscited by 108
  6. 2026EvoSkill: Automated Skill Discovery for Multi-Agent Systemscited by 96
  7. 2025WoW: Towards a World omniscient World model Through Embodied Interactioncited by 66
  8. 2026Memento-Skills: Let Agents Design Agentscited by 56
  9. 2025ThetaEvolve: Test-time Learning on Open Problemscited by 52
  10. 2025FLEX: Continuous Agent Evolution via Forward Learning from Experiencecited by 51
  11. 2026Expanding the Capabilities of Reinforcement Learning via Text Feedbackcited by 49
  12. 2025Barbarians at the Gate: How AI is Upending Systems Researchcited by 48
38 more
  1. 2025Astra: A Multi-Agent System for GPU Kernel Performance Optimizationcited by 46
  2. 2026AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimizationcited by 46
  3. 2026EvoX: Meta-Evolution for Automated Discoverycited by 45
  4. 2026Skill-Pro: Learning Reusable Skills from Experience via Non-Parametric PPO for LLM Agentscited by 44
  5. 2025Feedback Descent: Open-Ended Text Optimization via Pairwise Comparisoncited by 42
  6. 2026Meta Context Engineering via Agentic Skill Evolutioncited by 42
  7. 2026Code as Agent Harnesscited by 40
  8. 2026Rethinking the Evaluation of Harness Evolution for Agentscited by 39
  9. 2026CORAL: Towards Autonomous Multi-Agent Evolution for Open-Ended Discoverycited by 38
  10. 2025Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Modelscited by 30
  11. 2026Continual Harness: Online Adaptation for Self-Improving Foundation Agentscited by 30
  12. 2026Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agentscited by 28
  13. 2025Where Did It All Go Wrong? A Hierarchical Look into Multi-Agent Error Attributioncited by 25
  14. 2026Reasoning Models Struggle to Control their Chains of Thoughtcited by 22
  15. 2026From Static Templates to Dynamic Runtime Graphs: A Survey of Workflow Optimization for LLM Agentscited by 22
  16. 2026Hidden in Memory: Sleeper Memory Poisoning in LLM Agentscited by 22
  17. 2026Recursive Harness Self-Improvementcited by 22
  18. 2025AccelOpt: A Self-Improving LLM Agentic System for AI Accelerator Kernel Optimizationcited by 21
  19. 2026From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flawscited by 21
  20. 2025Speculative Actions: A Lossless Framework for Faster Agentic Systemscited by 20
  21. 2025FrontierCS: Evolving Challenges for Evolving Intelligencecited by 19
  22. 2025AutoEnv: Automated Environments for Measuring Cross-Environment Agent Learningcited by 18
  23. 2026RHO: Your Coding Agent is Secretly a Roboticistcited by 18
  24. 2025Trace Length is a Simple Uncertainty Signal in Reasoning Modelscited by 16
  25. 2025SCOPE: Prompt Evolution for Enhancing Agent Effectivenesscited by 16
  26. 2026Multi-Agent Teams Hold Experts Backcited by 16
  27. 2026Co-Harness: Co-Evolving Harnesses and Model Weights for LLM Agentscited by 16
  28. 2025REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understandingcited by 15
  29. 2026Model Context Protocol (MCP) Tool Descriptions Are Smelly! Towards Improving AI Agent Efficiency with Augmented MCP Tool Descriptionscited by 15
  30. 2026Evaluating Memory Structure in LLM Agentscited by 12
  31. 2026Autodata: An agentic data scientist to create high quality synthetic datacited by 12
  32. 2025Agentic Learner with Grow-and-Refine Multimodal Semantic Memorycited by 11
  33. 2026SkillAudit: Ground-Truth-Free Skill Evolution via Paired Trajectory Auditingcited by 11
  34. 2026HarnessBank: Semantic Gene-Bank Search with Gated Verification for Agent-Harness Self-Evolutioncited by 11
  35. 2026KernelBlaster: Continual Cross-Task CUDA Optimization via Memory-Augmented In-Context Reinforcement Learningcited by 10
  36. 2025Maestro: Joint Graph & Config Optimization for Reliable AI Agentscited by 9
  37. 2025Semantic Soft Bootstrapping: Long Context Reasoning in LLMs without Reinforcement Learningcited by 9
  38. 2026CreativeBench: Benchmarking and Enhancing Machine Creativity via Self-Evolving Challengescited by 9

50 most cited of 449 All 449 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs
  2. 2024Agent-Pro: Learning to Evolve via Policy-Level Reflection and Optimization
  3. 2026Skill-R1: Agent Skill Evolution via Reinforcement Learning
  4. 2026FLARE: Few-shot Learning-based Adaptive Reflective Engine
  5. 2025ToolExpander: Extending the Frontiers of Tool-Using Reinforcement Learning to Weak LLMs
  6. 2026Exploration-Driven Optimization for Test-Time Large Language Model Reasoning
  7. 2026Enhancing LLM-based Search Agents via Contribution Weighted Group Relative Policy Optimization
  8. 2024Entropy-Regularized Token-Level Policy Optimization for Language Agent Reinforcement