BaseReward: A Strong Baseline for Multimodal Reward Model

Yi-Fan Zhang; Haihua Yang; Huanyu Zhang; Yang Shi; Zezhou Chen; Haochen Tian; Chaoyou Fu; Haotian Wang; Kai Wu; Bo Cui; Xu Wang; Jianfei Pan; Haotian Wang; Zhang Zhang; Liang Wang

Paper · arXiv 2509.16127

BaseReward: A Strong Baseline for Multimodal Reward Model

Yi-Fan Zhang, Haihua Yang, Huanyu Zhang, Yang Shi, Zezhou Chen, Haochen Tian, Chaoyou Fu, Haotian Wang, Kai Wu, Bo Cui, Xu Wang, Jianfei Pan, Haotian Wang, Zhang Zhang, Liang Wang

21 upvotesSeptember 19, 2025arXiv 预印本

AI 摘要

The paper provides a comprehensive guide and introduces BaseReward, a state-of-the-art multimodal reward model, which outperforms existing models across various benchmarks and real-world tasks.

Multimodal Large Language ModelsReward ModelsMultimodal Reward Modelsreward modeling paradigmsreward head architecturetraining strategiesdata curationbackbone modelmodel scaleensemble methodsQwen2.5-VLMM-RLHF-Reward BenchVL-Reward BenchMultimodal Reward Benchreinforcement learning pipeline

Abstract

The rapid advancement of Multimodal Large Language Models (MLLMs) has made aligning them with human preferences a critical challenge. Reward Models (RMs) are a core technology for achieving this goal, but a systematic guide for building state-of-the-art Multimodal Reward Models (MRMs) is currently lacking in both academia and industry. Through exhaustive experimental analysis, this paper aims to provide a clear ``recipe'' for constructing high-performance MRMs. We systematically investigate every crucial component in the MRM development pipeline, including reward modeling paradigms (e.g., Naive-RM, Critic-based RM, and Generative RM), reward head architecture, training strategies, data curation (covering over ten multimodal and text-only preference datasets), backbone model and model scale, and ensemble methods. Based on these experimental insights, we introduce BaseReward, a powerful and efficient baseline for multimodal reward modeling. BaseReward adopts a simple yet effective architecture, built upon a {Qwen2.5-VL} backbone, featuring an optimized two-layer reward head, and is trained on a carefully curated mixture of high-quality multimodal and text-only preference data. Our results show that BaseReward establishes a new SOTA on major benchmarks such as MM-RLHF-Reward Bench, VL-Reward Bench, and Multimodal Reward Bench, outperforming previous models. Furthermore, to validate its practical utility beyond static benchmarks, we integrate BaseReward into a real-world reinforcement learning pipeline, successfully enhancing an MLLM's performance across various perception, reasoning, and conversational tasks. This work not only delivers a top-tier MRM but, more importantly, provides the community with a clear, empirically-backed guide for developing robust reward models for the next generation of MLLMs.