OpenVLThinkerV2: A Generalist Multimodal Reasoning Model for Multi-domain Visual Tasks
2026-04-09 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors address challenges in training open-source multimodal AI models that understand both images and language by introducing a new method called Gaussian GRPO (G²RPO). This method changes how rewards are processed during learning to make training more stable and fair across different tasks. They also add techniques to better balance simple visual understanding with deeper reasoning. Using these improvements, they build OpenVLThinkerV2, a model that performs well across many tests compared to other top models. Their work helps improve how multimodal AI learns from diverse tasks.
Reinforcement LearningMultimodal ModelsPolicy OptimizationReward DistributionAdvantage FunctionGradient EquityEntropy ShapingVisual GroundingReasoning ChainsMultimodal Benchmarks
Authors
Wenbo Hu, Xin Chen, Yan Gao-Tian, Yihe Deng, Nanyun Peng, Kai-Wei Chang
Abstract
Group Relative Policy Optimization (GRPO) has emerged as the de facto Reinforcement Learning (RL) objective driving recent advancements in Multimodal Large Language Models. However, extending this success to open-source multimodal generalist models remains heavily constrained by two primary challenges: the extreme variance in reward topologies across diverse visual tasks, and the inherent difficulty of balancing fine-grained perception with multi-step reasoning capabilities. To address these issues, we introduce Gaussian GRPO (G$^2$RPO), a novel RL training objective that replaces standard linear scaling with non-linear distributional matching. By mathematically forcing the advantage distribution of any given task to strictly converge to a standard normal distribution, $\mathcal{N}(0,1)$, G$^2$RPO theoretically ensures inter-task gradient equity, mitigates vulnerabilities to heavy-tail outliers, and offers symmetric update for positive and negative rewards. Leveraging the enhanced training stability provided by G$^2$RPO, we introduce two task-level shaping mechanisms to seamlessly balance perception and reasoning. First, response length shaping dynamically elicits extended reasoning chains for complex queries while enforce direct outputs to bolster visual grounding. Second, entropy shaping tightly bounds the model's exploration zone, effectively preventing both entropy collapse and entropy explosion. Integrating these methodologies, we present OpenVLThinkerV2, a highly robust, general-purpose multimodal model. Extensive evaluations across 18 diverse benchmarks demonstrate its superior performance over strong open-source and leading proprietary frontier models.