Research map / Reinforcement learning / RL from feedback
GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
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 and 5 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use GEPA reflective prompt evolution instead of GRPO (24,000 rollouts) | Qwen3-8B · six tasks aggregate | score: 48.91 -> 54.85 | Improved |
| Use GEPA instead of GRPO | Qwen3-8B · AIME-2025 | accuracy: 38.00 -> 32.00 | Hurt |
| Use GEPA instead of MIPROv2 | Qwen3-8B · aggregate | score: 47.84 -> 54.85 | Improved |
| Use GEPA instead of MIPROv2 | GPT-4.1 Mini · aggregate | score: 58.67 -> 65.22 | Improved |
| Pareto-based candidate selection instead of SelectBestCandidate (ablation) | GEPA · Qwen3-8B 4-task aggregate | score: 54.89 -> 61.28 | Improved |
| Pareto-based candidate selection instead of BeamSearch (ablation) | GEPA · Qwen3-8B 4-task aggregate | score: 53.95 -> 61.28 | Improved |
| Add Merge (system-aware crossover) on GPT-4.1 Mini (ablation) | GEPA · GPT-4.1 Mini aggregate | score: 65.22 -> 66.36 | Improved |
| Add Merge (system-aware crossover) on Qwen3-8B (ablation) | GEPA · Qwen3-8B aggregate | score: 54.85 -> 52.40 | Hurt |
| Transfer prompts optimized on Qwen3-8B to GPT-4.1 Mini | GEPA · GPT-4.1 Mini aggregate | score: 53.03 -> 62.03 | Improved |
| Use GEPA instead of MIPROv2 prompt style (instruction-only) | GEPA · prompt length | tokens | Improved |
| Apply GEPA to Sequential10 NPU kernel agent | GPT-4o · NPUEval | mean vector utilization: 4.25% -> 30.52% | Improved |
Closest papers by meaning
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.89)
- Reward-Guided Prompt Evolving in Reinforcement Learning for LLMs (ICML 2025 · similarity 0.89)
- ESSAM: A Novel Competitive Evolution Strategies Approach to Reinforcement Learning for Memory Efficient LLMs Fine-Tuning (NeurIPS 2026 · similarity 0.88)
- Retroformer: Retrospective Large Language Agents with Policy Gradient Optimization (ICLR 2024 · similarity 0.88)
- ToolSearcher: Optimizing Tool Selection at Scale via Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Advancing LLM Reasoning with Natural Language and Numerical Feedback (ICML 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
