ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning

• URL: http://arxiv.org/abs/2512.05111v1
• Date: Thu, 04 Dec 2025 18:59:52 GMT
• Title: ARM-Thinker: Reinforcing Multimodal Generative Reward Models with Agentic Tool Use and Visual Reasoning
• Authors: Shengyuan Ding, Xinyu Fang, Ziyu Liu, Yuhang Zang, Yuhang Cao, Xiangyu Zhao, Haodong Duan, Xiaoyi Dong, Jianze Liang, Bin Wang, Conghui He, Dahua Lin, Jiaqi Wang,
• Abstract summary: ARM-Thinker is an Agentic multimodal Reward Model that autonomously invokes external tools to ground judgments in verifiable evidence.<n>We train ARM-Thinker with multi-stage reinforcement learning, jointly optimizing tool-calling decisions and judgment accuracy.<n>Our results demonstrate that agentic capabilities significantly enhance both accuracy and interpretability of reward models.
• Score: 103.7657839292775
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Reward models are critical for aligning vision-language systems with human preferences, yet current approaches suffer from hallucination, weak visual grounding, and an inability to use tools for verification, limiting their reliability on complex multimodal reasoning tasks. We present ARM-Thinker, an A}gentic multimodal Reward Model that autonomously invokes external tools (e.g., image cropping, doc page retrieval) to ground judgments in verifiable evidence, replacing static, non-interactive reward scoring. This enables the model to verify fine-grained visual details, cross-reference multi-page evidence, and validate reasoning claims, which are capabilities absent in existing reward models. We train ARM-Thinker with multi-stage reinforcement learning, jointly optimizing tool-calling decisions and judgment accuracy. To evaluate agentic reward modeling, we introduce ARMBench-VL, comprising three benchmarks that assess fine-grained visual grounding (image-level tools), multi-page document understanding (retrieval tools), and instruction following (text-level verification). ARM-Thinker achieves +16.2% average improvement on reward modeling benchmarks, +9.6% on tool-use tasks, and outperforms baselines on multimodal math and logical reasoning benchmarks. Our results demonstrate that agentic capabilities significantly enhance both accuracy and interpretability of reward models.

Related papers

• MSVBench: Towards Human-Level Evaluation of Multi-Shot Video Generation [48.84450712826316]
  MSVBench is the first comprehensive benchmark featuring hierarchical scripts and reference images tailored for Multi-Shot Video generation.<n>We propose a hybrid evaluation framework that synergizes the high-level semantic reasoning of Large Multimodal Models with the fine-grained perceptual rigor of domain-specific expert models.
  arXiv  Detail & Related papers  (2026-02-27T12:26:34Z)
• CoSineVerifier: Tool-Augmented Answer Verification for Computation-Oriented Scientific Questions [32.14674040685995]
  We introduce model, a tool-augmented verifier that leverages external rubrics to perform precise computations and symbolic simplifications.<n>Experiments conducted on STEM subjects, general QA, and long-form reasoning tasks demonstrates strong generalization of model.
  arXiv  Detail & Related papers  (2025-12-01T03:08:43Z)
• TIR-Bench: A Comprehensive Benchmark for Agentic Thinking-with-Images Reasoning [30.018325742295243]
  OpenAI o3 can create and operate tools to transform images for problem-solving, also known as thinking-textitwith-images in chain-of-thought.<n>Visual Search tests only basic operations such as localization and cropping, offering little insight into more complex, dynamic, and tool-dependent reasoning.<n>We introduce textbfTIR-Bench, a comprehensive benchmark for evaluating agentic thinking-with-images across 13 diverse tasks.
  arXiv  Detail & Related papers  (2025-11-03T18:40:17Z)
• One Model to Critique Them All: Rewarding Agentic Tool-Use via Efficient Reasoning [54.580646706013965]
  Reward models (RMs) play a critical role in aligning large language models with human preferences.<n>We introduce ToolRM, a family of lightweight generative RMs tailored for general tool-use scenarios.<n>To build these models, we propose a novel pipeline that constructs pairwise preference data using rule-based scoring and multidimensional sampling.
  arXiv  Detail & Related papers  (2025-10-30T06:08:27Z)
• On Generalization in Agentic Tool Calling: CoreThink Agentic Reasoner and MAVEN Dataset [16.921428284844684]
  Generalization across Agentic tool-calling environments remains a key unsolved challenge in developing reliable reasoning systems.<n>We present a framework that augments large language models with a lightweight symbolic reasoning layer for structured decomposition and adaptive tool orchestration.
  arXiv  Detail & Related papers  (2025-10-27T00:58:48Z)
• Self-Consistency as a Free Lunch: Reducing Hallucinations in Vision-Language Models via Self-Reflection [71.8243083897721]
  Vision-language models often hallucinate details, generating non-existent objects or inaccurate attributes that compromise output reliability.<n>We present a novel framework that leverages the model's self-consistency between long responses and short answers to generate preference pairs for training.
  arXiv  Detail & Related papers  (2025-09-27T10:37:11Z)
• VisualPRM: An Effective Process Reward Model for Multimodal Reasoning [76.35753243272521]
  We introduce VisualPRM, which improves the reasoning abilities of existing Multimodal Large Language Models (MLLMs)<n>Our model achieves a 5.9-point improvement across seven multimodal reasoning benchmarks.<n>For the evaluation of multimodal PRMs, we propose VisualProcessBench, a benchmark with human-annotated step-wise correctness labels.
  arXiv  Detail & Related papers  (2025-03-13T12:03:37Z)
• ToolComp: A Multi-Tool Reasoning & Process Supervision Benchmark [0.0]
  We introduce ToolComp, a benchmark designed to evaluate multi-step tool-use reasoning.<n>ToolComp is developed through a collaboration between models and human annotators.<n>We generate synthetic training data to compare the performance of outcome-supervised reward models with process-supervised reward models.
  arXiv  Detail & Related papers  (2025-01-02T15:10:52Z)
• RewardBench: Evaluating Reward Models for Language Modeling [100.28366840977966]
  We present RewardBench, a benchmark dataset and code-base for evaluation of reward models. The dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods.
  arXiv  Detail & Related papers  (2024-03-20T17:49:54Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.