Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning

Xinyu Tang, Gangqiang Cao, Yurou Liu, Yuliang Zhan, Xiaochong Lan, Yifan Li, Yuchen Yan, Han Peng, Zican Dong, Zhenduo Zhang, Tianshu Wang, Xinyu Kong, Zujie Wen, Wayne Xin Zhao, Zhiqiang Zhang, Jun Zhou

arXiv:2607.12395 · 2026-07-16 공개 · arXiv · PDF

chain-of-thought model-scaling mathematical-benchmarks structured-evaluation training-optimization self-verification zero-rl large-scale-rl

Abstract

Reinforcement learning with verifiable rewards without human-annotated data, often referred to as zero RL, has emerged as a powerful paradigm for eliciting chain-of-thought reasoning. However, due to computational constraints, existing studies are largely restricted to small models, leaving the training dynamics and emergent capabilities at a large scale unexplored. To meaningfully explore this frontier, we aim to elicit high-quality reasoning behaviors from the model. However, we find that naive scaling often suffers from poor readability, token redundancy, and a lack of adaptive reasoning depth. To address these challenges, we present a stable and efficient training pipeline, incorporating algorithmic and system optimizations such as clipped importance sampling, training-inference ratio correction, and mixed-precision control. Our experiments offer three key findings that validate the "bitter lesson" of scaling: (1) scaling to 1T parameters significantly enhances sample efficiency and performance ceilings; (2) the training process progresses sequentially through an initial discovery phase followed by a sharpening phase; and (3) the model spontaneously develops advanced cognitive behaviors, including anthropomorphism, structured formatting, self-verification, parallel reasoning, and context anxiety, rendering hand-crafted heuristics redundant. Evaluated on seven mathematical benchmarks, Ring-2.5-1T-Zero achieves competitive performance. Additionally, to assess CoT quality beyond final-answer correctness, we propose a structured evaluation framework across three dimensions: comprehensibility, reproducibility, and efficiency, where our model demonstrates clear advantages in producing structured and concise reasoning traces. By sharing our observed emergent phenomena, we hope to provide the community with deeper insights into scaling behaviors, particularly at the 1-trillion scale.

한국어 요약

한 줄 요약

1조 파라미터 규모의 Ring-2.5-1T-Zero 모델을 통해 Zero RL 기반 추론 능력이 자발적으로 발현되며, 기존 휴리스틱이 불필요해짐.

핵심 기여도

핵심 아이디어

Zero RL은 인간 라벨 없이도 추론 능력을 유도할 수 있지만, 기존 연구는 소규모 모델에 제한되어 있었다. 본 연구는 1조 파라미터 규모에서 Zero RL을 적용하여, 추론 능력의 자발적 발현과 학습 역학을 탐구했다. 기존 Zero RL의 단점인 읽기 어려움, 토큰 중복, 고정된 추론 깊이를 해결하기 위해, clipped importance sampling, training-inference ratio correction, mixed-precision control 등의 최적화 기법을 도입했다. 학습 단계는 "discovery phase"와 "sharpening phase"로 구분되며, 이는 모델이 추론 경계를 확장한 후 정제하는 과정을 반영한다. 학습 초기에는 토큰 수가 급증하는 문제가 있었으나, sample-level loss normalization과 tier-based adaptive training을 통해 이 문제를 해결했다.

기술적 접근법

주요 결과

의의 및 한계

실용적 활용