Exploring Reasoning Reward Model for Agents

Agentic Reinforcement Learning (Agentic RL) has achieved notable success in enabling agents to perform complex reasoning and tool use. However, most methods still relies on sparse outcome-based reward for training. Such feedback fails to differentiate intermediate reasoning quality, leading to suboptimal training results. In this paper, we introduce Agent Reasoning Reward Model (Agent-RRM), a multi-faceted reward model that produces structured feedback for agentic trajectories, including (1) an explicit reasoning trace , (2) a focused critique that provides refinement guidance by highlighting reasoning flaws, and (3) an overall score that evaluates process performance. Leveraging these signals, we systematically investigate three integration strategies: Reagent-C (text-augmented refinement), Reagent-R (reward-augmented guidance), and Reagent-U (unified feedback integration). Extensive evaluations across 12 diverse benchmarks demonstrate that Reagent-U yields substantial performance leaps, achieving 43.7% on GAIA and 46.2% on WebWalkerQA, validating the effectiveness of our reasoning reward model and training schemes. Code, models, and datasets are all released to facilitate future research.

💡 Research Summary

The paper tackles a fundamental limitation of current agentic reinforcement learning (RL) systems: the reliance on sparse, outcome‑based rewards that only evaluate the final answer. Such binary feedback cannot distinguish between high‑quality intermediate reasoning steps and completely erroneous attempts, leading to sub‑optimal policy learning, especially in long‑horizon tasks that involve tool use and multi‑step reasoning. To address this, the authors introduce the Agent Reasoning Reward Model (Agent‑RRM), a multi‑faceted evaluator that generates three structured signals for each agent trajectory: (1) a reasoning trace that explicitly analyzes logical consistency, (2) a that pinpoints specific reasoning or execution flaws, and (3) a scalar in the range