Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoning

TL;DR

This paper introduces Open-Vision-Reasoner, a multimodal large language model trained through a two-stage process combining linguistic fine-tuning and reinforcement learning, achieving state-of-the-art visual reasoning performance.

Contribution

It presents a novel two-stage training paradigm for MLLMs that transfers cognitive behaviors from language models to enhance visual reasoning capabilities.

Findings

Behavior transfer occurs early due to linguistic mental imagery.

Cold start memorizes visual behaviors; RL scales effective patterns.

Transfer emphasizes high-utility behaviors like visual reflection.

Abstract

The remarkable reasoning capability of large language models (LLMs) stems from cognitive behaviors that emerge through reinforcement with verifiable rewards. This work investigates how to transfer this principle to Multimodal LLMs (MLLMs) to unlock advanced visual reasoning. We introduce a two-stage paradigm built on Qwen2.5-VL-7B: a massive linguistic cold-start fine-tuning, followed by multimodal reinforcement learning (RL) spanning nearly 1,000 steps, surpassing all previous open-source efforts in scale. This pioneering work reveals three fundamental insights: 1) Behavior transfer emerges surprisingly early in cold start due to linguistic mental imagery. 2) Cold start broadly memorizes visual behaviors, while RL critically discerns and scales up effective patterns. 3) Transfer strategically favors high-utility behaviors such as visual reflection. Our resulting model, Open-Vision-Reasoner (OVR), achieves state-of-the-art performance on a suite of reasoning benchmarks, including 95.3% on MATH500, 51.8% on MathVision and 54.6% on MathVerse. We release our model, data, and training dynamics to catalyze the development of more capable, behavior-aligned multimodal reasoners.