Saved in:
Bibliographic Details
Main Authors: Shang, Ning, Liu, Yifei, Zhu, Yi, Zhang, Li Lyna, Xu, Weijiang, Guan, Xinyu, Zhang, Buze, Dong, Bingcheng, Zhou, Xudong, Zhang, Bowen, Xin, Ying, Miao, Ziming, Li, Scarlett, Yang, Fan, Yang, Mao
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.20722
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908508229730304
author Shang, Ning
Liu, Yifei
Zhu, Yi
Zhang, Li Lyna
Xu, Weijiang
Guan, Xinyu
Zhang, Buze
Dong, Bingcheng
Zhou, Xudong
Zhang, Bowen
Xin, Ying
Miao, Ziming
Li, Scarlett
Yang, Fan
Yang, Mao
author_facet Shang, Ning
Liu, Yifei
Zhu, Yi
Zhang, Li Lyna
Xu, Weijiang
Guan, Xinyu
Zhang, Buze
Dong, Bingcheng
Zhou, Xudong
Zhang, Bowen
Xin, Ying
Miao, Ziming
Li, Scarlett
Yang, Fan
Yang, Mao
contents We introduce rStar2-Agent, a 14B math reasoning model trained with agentic reinforcement learning to achieve frontier-level performance. Beyond current long CoT, the model demonstrates advanced cognitive behaviors, such as thinking carefully before using Python coding tools and reflecting on code execution feedback to autonomously explore, verify, and refine intermediate steps in complex problem-solving. This capability is enabled through three key innovations that makes agentic RL effective at scale: (i) an efficient RL infrastructure with a reliable Python code environment that supports high-throughput execution and mitigates the high rollout costs, enabling training on limited GPU resources (64 MI300X GPUs); (ii) GRPO-RoC, an agentic RL algorithm with a Resample-on-Correct rollout strategy that addresses the inherent environment noises from coding tools, allowing the model to reason more effectively in a code environment; (iii) An efficient agent training recipe that starts with non-reasoning SFT and progresses through multi-RL stages, yielding advanced cognitive abilities with minimal compute cost. To this end, rStar2-Agent boosts a pre-trained 14B model to state of the art in only 510 RL steps within one week, achieving average pass@1 scores of 80.6% on AIME24 and 69.8% on AIME25, surpassing DeepSeek-R1 (671B) with significantly shorter responses. Beyond mathematics, rStar2-Agent-14B also demonstrates strong generalization to alignment, scientific reasoning, and agentic tool-use tasks. Code and training recipes are available at https://github.com/microsoft/rStar.
format Preprint
id arxiv_https___arxiv_org_abs_2508_20722
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle rStar2-Agent: Agentic Reasoning Technical Report
Shang, Ning
Liu, Yifei
Zhu, Yi
Zhang, Li Lyna
Xu, Weijiang
Guan, Xinyu
Zhang, Buze
Dong, Bingcheng
Zhou, Xudong
Zhang, Bowen
Xin, Ying
Miao, Ziming
Li, Scarlett
Yang, Fan
Yang, Mao
Computation and Language
We introduce rStar2-Agent, a 14B math reasoning model trained with agentic reinforcement learning to achieve frontier-level performance. Beyond current long CoT, the model demonstrates advanced cognitive behaviors, such as thinking carefully before using Python coding tools and reflecting on code execution feedback to autonomously explore, verify, and refine intermediate steps in complex problem-solving. This capability is enabled through three key innovations that makes agentic RL effective at scale: (i) an efficient RL infrastructure with a reliable Python code environment that supports high-throughput execution and mitigates the high rollout costs, enabling training on limited GPU resources (64 MI300X GPUs); (ii) GRPO-RoC, an agentic RL algorithm with a Resample-on-Correct rollout strategy that addresses the inherent environment noises from coding tools, allowing the model to reason more effectively in a code environment; (iii) An efficient agent training recipe that starts with non-reasoning SFT and progresses through multi-RL stages, yielding advanced cognitive abilities with minimal compute cost. To this end, rStar2-Agent boosts a pre-trained 14B model to state of the art in only 510 RL steps within one week, achieving average pass@1 scores of 80.6% on AIME24 and 69.8% on AIME25, surpassing DeepSeek-R1 (671B) with significantly shorter responses. Beyond mathematics, rStar2-Agent-14B also demonstrates strong generalization to alignment, scientific reasoning, and agentic tool-use tasks. Code and training recipes are available at https://github.com/microsoft/rStar.
title rStar2-Agent: Agentic Reasoning Technical Report
topic Computation and Language
url https://arxiv.org/abs/2508.20722