UloRL:An Ultra-Long Output Reinforcement Learning Approach for Advancing Large Language Models' Reasoning Abilities

TL;DR

This paper introduces UloRL, a novel reinforcement learning method that enhances large language models' reasoning by efficiently training on ultra-long outputs through segmentation and dynamic masking, leading to significant performance improvements.

Contribution

We propose UloRL, which segments ultra-long outputs and employs dynamic masking to improve training efficiency and reasoning abilities of large language models.

Findings

2.06x faster training speed on Qwen3-30B-A3B

Performance improvement on AIME2025 from 70.9% to 85.1%

Outperforms larger models with notable gains

Abstract

Recent advances in large language models (LLMs) have highlighted the potential of reinforcement learning with verifiable rewards (RLVR) to enhance reasoning capabilities through extended output sequences. However, traditional RL frameworks face inefficiencies when handling ultra-long outputs due to long-tail sequence distributions and entropy collapse during training. To address these challenges, we propose an Ultra-Long Output Reinforcement Learning (UloRL) approach for advancing large language models' reasoning abilities. Specifically, we divide ultra long output decoding into short segments, enabling efficient training by mitigating delays caused by long-tail samples. Additionally, we introduce dynamic masking of well-Mastered Positive Tokens (MPTs) to prevent entropy collapse. Experimental results demonstrate the effectiveness of our approach. On the Qwen3-30B-A3B model, RL with segment rollout achieved 2.06x increase in training speed, while RL training with 128k-token outputs improves the model's performance on AIME2025 from 70.9\% to 85.1\% and on BeyondAIME from 50.7\% to 61.9\%, even surpassing Qwen3-235B-A22B with remarkable gains. These findings underscore the potential of our methods to advance the reasoning capabilities of LLMs with ultra-long sequence generation. We will release our code and model for further use by the community.