SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks

Computer Science > Artificial Intelligence [Submitted on 10 Apr 2026] SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks Tianyi Wang, Yixia Li, Long Li, Yibiao Chen, Shaohan Huang, Yun Chen, Peng Li, Yang Liu, Guanhua Chen Proximal Policy Optimization (PPO) is central to aligning Large Language Models (LLMs) in reasoning tasks with verifiable rewards. However, standard token-level PPO struggles in this setting due to the instability of temporal credit assignment over long Chain-of-Thought (CoT) horizons and the prohibitive memory cost of the value model. While critic-free alternatives like GRPO mitigate these issues, they incur significant computational overhead by requiring multiple samples for baseline estimation, severely limiting training throughput. In this paper, we introduce Sequence-Level PPO (SPPO), a scalable algorithm that harmonizes the sample efficiency of PPO with the stability of outcome-based updates. SPPO reformulates the reasoning process as a Sequence-Level Contextual Bandit problem, employing a decoupled scalar value function to derive low-variance advantage signals without multi-sampling. Extensive experiments on mathematical benchmarks demonstrate that SPPO significantly surpasses standard PPO and matches the performance of computation-heavy group-based methods, offering a resource-efficient framework for aligning reasoning LLMs. Comments: ACL 2026 Main Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2604.08865 [cs.AI]   (or arXiv:2604.08865v1 [cs.AI] for this version)   https://doi.org/10.48550/arXiv.2604.08865 Focus to learn more Submission history From: Tianyi Wang [view email] [v1] Fri, 10 Apr 2026 01:58:21 UTC (5,501 KB) Access Paper: HTML (experimental) view license Current browse context: cs.AI < prev   |   next > new | recent | 2026-04 Change to browse by: cs References & Citations NASA ADS Google Scholar Semantic Scholar Export BibTeX Citation Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Demos Related Papers About arXivLabs Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)