ModelRefs / GRPO (Group Relative Policy Optimization) — AI Glossary

GRPO (Group Relative Policy Optimization) — AI Glossary

A reinforcement learning algorithm for LLMs that replaces the critic model with group-based advantage estimation, used in DeepSeek-R1.

Overview

GRPO (Shao et al. 2024) generates a group of completions per prompt, uses their relative rewards to estimate advantages (no separate critic model needed), and applies PPO-style clipping. Simpler and more compute-efficient than PPO for LLM fine-tuning. Used in DeepSeek-R1 and Qwen reasoning models to achieve o1-level math performance.

Reference details

Topictraining
Last reviewed2026-06-24

Primary source

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to GRPO (Group Relative Policy Optimization) — AI Glossary.

Frequently asked questions

What is GRPO (Group Relative Policy Optimization)?

A reinforcement learning algorithm for LLMs that replaces the critic model with group-based advantage estimation, used in DeepSeek-R1.

What concepts are related to GRPO (Group Relative Policy Optimization)?

Closely related concepts include ppo, rlhf, verifiable reward.