OpenVLThinkerV2: A Generalist Multimodal Reasoning Model for Multi-domain Visual Tasks

TL;DR

OpenVLThinkerV2 is a versatile multimodal reasoning model that employs a novel RL training objective, G$^2$RPO, to improve multi-domain visual task performance through balanced perception and reasoning.

Contribution

The paper introduces G$^2$RPO, a distributional matching RL objective, and task-level shaping mechanisms, enabling robust, general-purpose multimodal models for diverse visual tasks.

Findings

Outperforms strong open-source and proprietary models on 18 benchmarks.

G$^2$RPO improves training stability and inter-task gradient equity.

Task shaping mechanisms effectively balance perception and reasoning.

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.