Understanding the Role of Hallucination in Reinforcement Post-Training of Multimodal Reasoning Models

TL;DR

This paper introduces the Hallucination-as-Cue Framework to analyze how reinforcement learning post-training influences multimodal reasoning models, revealing that hallucination plays a significant role in their improved performance.

Contribution

It proposes a novel analytical framework to diagnose RL training effects on multimodal models by using hallucination-inductive corruptions, challenging existing assumptions.

Findings

RL post-training can improve reasoning even with hallucination-inductive corruptions

Models trained with RL outperform standard training in some benchmarks

Hallucination significantly influences the reasoning capabilities of multimodal models

Abstract

The recent success of reinforcement learning (RL) in large reasoning models has inspired the growing adoption of RL for post-training Multimodal Large Language Models (MLLMs) to enhance their visual reasoning capabilities. Although many studies have reported improved performance, it remains unclear whether RL training truly enables models to learn from visual information. In this work, we propose the Hallucination-as-Cue Framework, an analytical framework designed to investigate the effects of RL-based post-training on multimodal reasoning models from the perspective of model hallucination. Specifically, we introduce hallucination-inductive, modality-specific corruptions that remove or replace essential information required to derive correct answers, thereby forcing the model to reason by hallucination. By applying these corruptions during both training and evaluation, our framework provides a unique perspective for diagnosing RL training dynamics and understanding the intrinsic properties of datasets. Through extensive experiments and analyses across multiple multimodal reasoning benchmarks, we reveal that the role of model hallucination for RL-training is more significant than previously recognized. For instance, we find that RL post-training under purely hallucination-inductive settings can still significantly improve models' reasoning performance, and in some cases even outperform standard training. These findings challenge prevailing assumptions about MLLM reasoning training and motivate the development of more modality-aware RL-based training designs.