# M-GRPO: Stabilizing Self-Supervised Reinforcement Learning for Large Language Models with Momentum-Anchored Policy Optimization

arXiv: https://arxiv.org/abs/2512.13070 · HTML: https://mlsanity.com/paper/2512.13070/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning.
- Builds on 13 of its 14 references that are in the map.
- First posted on arXiv 15 Dec 2025.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)

## Builds on (most cited first, 13 of 13)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Momentum Contrast for Unsupervised Visual Representation Learning](https://mlsanity.com/paper/1911.05722/) (2019)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Improved Baselines with Momentum Contrastive Learning](https://mlsanity.com/paper/2003.04297/) (2020)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models](https://mlsanity.com/paper/2505.22617/) (2025)
- [TTRL: Test-Time Reinforcement Learning](https://mlsanity.com/paper/2504.16084/) (2025)
- [Learning to Reason without External Rewards](https://mlsanity.com/paper/2505.19590/) (2025)
- [EDGE-GRPO: Entropy-Driven GRPO with Guided Error Correction for Advantage Diversity](https://mlsanity.com/paper/2507.21848/) (2025)
- [Scaling Behaviors of LLM Reinforcement Learning Post-Training: An Empirical Study in Mathematical Reasoning](https://mlsanity.com/paper/2509.25300/) (2025)

## Built on by (most cited first, 4 of 4)

- [Advantage Collapse in Group Relative Policy Optimization: Diagnosis and Mitigation](https://mlsanity.com/paper/2605.21125/) (2026)
- [Learning from the Self-future: On-policy Self-distillation for dLLMs](https://mlsanity.com/paper/2606.18195/) (2026)
- [Demystifying Design Choices of Reinforcement Fine-tuning: A Batched Contextual Bandit Learning Perspective](https://mlsanity.com/paper/2601.22532/) (2026)
- [Stable-MM-R1: Anchoring Multimodal Reasoning Dynamics via Entropy-Guided Stratification](https://mlsanity.com/paper/2609.07148/) (2026)

## Abstract (arXiv)

Self-supervised reinforcement learning (RL) presents a promising approach for enhancing the reasoning capabilities of Large Language Models (LLMs) without reliance on expensive human-annotated data. However, we find that existing methods suffer from a critical failure mode under long-horizon training: a "policy collapse" where performance precipitously degrades. We diagnose this instability and demonstrate that simply scaling the number of rollouts -- a common strategy to improve performance -- only delays, but does not prevent, this collapse. To counteract this instability, we first introduce M-GRPO (Momentum-Anchored Group Relative Policy Optimization), a framework that leverages a slowly evolving momentum model to provide a stable training target. In addition, we identify that this process is often accompanied by a rapid collapse in policy entropy, resulting in a prematurely confident and suboptimal policy. To specifically address this issue, we propose a second contribution: an adaptive filtering method based on the interquartile range (IQR) that dynamically prunes low-entropy trajectories, preserving essential policy diversity. Our extensive experiments on multiple reasoning benchmarks demonstrate that M-GRPO stabilizes the training process while the IQR filter prevents premature convergence. The combination of these two innovations leads to superior training stability and state-of-the-art performance.