Organizations: Department of Electrical and Computer Engineering, University of Washington, Seattle, WA, USA · The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China · Department of Electrical Engineering, City University of Hong Kong, Hong Kong SAR, China
Stabilizing long-horizon reasoning requires intelligent agents to sustain useful progress across successive reasoning steps and tool interactions. Such agents are typically trained with reinforcement learning (RL), where actor-critic RL uses a learnable critic network to estimate step advantages, while critic-free methods such as GRPO eliminate critic models and rely mostly on the group variants of sparse outcome-only rewards. Both reward types leave partially useful failed trajectories either implicit or indistinguishable from entirely unproductive ones, and as reasoning horizons grow, it becomes difficult to maintain consistent reasoning trajectories. In this work, we propose a plug-and-play verifier framework for stabilizing long-horizon reasoning through explicit verification of intermediate reasoning artifacts. Specifically, the framework decouples process verification through a unified interface that extracts intermediate reasoning artifacts from agent running trajectories, evaluates them using domain-specific verifiers, and aggregates the verification scores as bounded process reward. Then, we combine our dense verified process rewards and the sparse outcome rewards through a two-stage post-training schedule. We showed that with plug-and-play verifiers, we further stabilize the reasoning steps as last-mile self-reflection and self-consistency. Typical evaluation on HotpotQA using Qwen3.5-4B achieves a superior accuracy (60.18%), outperforming the standard GRPO baseline (57.73%). Our results suggest that plug-and-play verifiers provide a modular approach to learn more stable and reliable reasoning behaviors over long horizons.
Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning capabilities of large language models (LLMs), yet existing approaches predominantly rely on sparse outcome-level feedback, posing a substantial credit-assignment challenge in long-horizon agentic tasks. In this work, we study structured agentic reasoning problems that admit symbolic or algorithmic task-solving oracles. We propose Verifiable Process Rewards (VPR), a training framework that repurposes such task oracles as action-level process verifiers, providing dense rewards for intermediate actions and enabling fine-grained credit assignment. We instantiate VPR in three distinct settings: a search-based oracle for Sokoban, a constraint-based oracle for Sudoku, and a posterior-based oracle for Minesweeper. To optimize policies with these rewards, we introduce oracle-guided state-group rollout and locally normalized optimization. Empirically, VPR-trained Qwen3-4B outperforms outcome-reward GRPO and VinePPO across all three training environments. When jointly trained on math data and all three game environments, VPR achieves the highest average general-reasoning performance and strongest agentic results among the compared training configurations. To further assess its applicability to more general agentic tool-use settings, we apply VPR-style action-level supervision to τ2-bench using an expert LLM as a reference policy and observe improved performance over outcome-reward GRPO on held-out tasks. Together, these results position VPR as a promising approach for leveraging reliable task oracles to improve agentic reasoning, while highlighting its dependence on the availability and reliability of task oracles and the challenge of extending intermediate verification to less structured, open-ended environments.
Reasoning in knowledge-intensive domains remains challenging as intermediate steps are often not locally verifiable: unlike math or code, evaluating step correctness may require synthesizing clues across large external knowledge sources. As a result, subtle errors can propagate through reasoning traces, potentially never to be detected. Prior work has proposed process reward models (PRMs), including retrieval-augmented variants, but these methods operate post hoc, scoring completed trajectories, which prevents their integration into dynamic inference procedures. Here, we introduce Process Reward Agents (PRA), an inference-time method for providing domain-grounded, online, step-wise rewards to a frozen policy. In contrast to prior retrieval-augmented PRMs, PRA enables search-based decoding to rank and prune candidate trajectories at every generation step. Experiments on multiple medical reasoning benchmarks demonstrate that PRA consistently outperforms strong baselines, achieving 81.9% accuracy on MedQA with Qwen3-4B, a new state of the art at the 4B scale. Importantly, PRA generalizes to unseen frozen policy models ranging from 0.5B to 8B parameters, improving their accuracy by up to 25.7% without any policy model updates. More broadly, PRA suggests a paradigm in which frozen reasoners are decoupled from domain-specific reward modules, allowing the deployment of new backbones in complex domains without retraining.
Jiwoong Sohn, Tomasz Sternal, Kenneth Styppa +2
ETH Zürich, Switzerland · Heidelberg University, Germany
Verifiers have been demonstrated to enhance LLM reasoning via test-time scaling (TTS). Yet, they face significant challenges in complex domains. Error propagation from incorrect intermediate reasoning can lead to false positives for seemingly plausible solutions, while lacking external grounding makes verifiers unreliable on computation or knowledge-intensive tasks. To address these challenges, we propose Agentic Verifier, a framework that transforms reward modeling into a multi-turn, tool-augmented deliberative process. We introduce complementary forward and backward agents: one traces solutions from premises to conclusions, while the other re-checks conclusions against their underlying premises. This bidirectional process enables a comprehensive, reliable, and interpretable assessment of solutions. To facilitate practical deployment, we propose AgentV-RL. Through proactive exploration and reinforcement learning, the verifier autonomously interleaves tool-use with internal reasoning. Extensive experiments show that Agentic Verifier yields consistent performance gains under both parallel and sequential TTS. Notably, our 4B variant surpasses state-of-the-art ORMs by 25.2%, positioning it as a promising paradigm for agentic reward modeling.
Jiazheng Zhang, Ziche Fu, Zhiheng Xi +13
College of Computer Science and Artificial Intelligence, Fudan University · Huazhong University of Science and Technology · The University of Hong Kong +3