# Do Vision-Language Models Truly Perform Vision Reasoning? A Rigorous Study of the Modality Gap

arXiv: https://arxiv.org/abs/2604.16256 · HTML: https://mlsanity.com/paper/2604.16256/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 15 of its 42 references that are in the map.
- First posted on arXiv 17 Apr 2026.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 15 of 15)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [OpenAI GPT-5 System Card](https://mlsanity.com/paper/2601.03267/) (2025)
- [VisualPuzzles: Decoupling Multimodal Reasoning Evaluation from Domain Knowledge](https://mlsanity.com/paper/2504.10342/) (2025)
- [SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought Reasoning](https://mlsanity.com/paper/2505.11484/) (2025)
- [Mind the Gap: Benchmarking Spatial Reasoning in Vision-Language Models](https://mlsanity.com/paper/2503.19707/) (2025)
- [VGRP-Bench: Visual Grid Reasoning Puzzle Benchmark for Large Vision-Language Models](https://mlsanity.com/paper/2503.23064/) (2025)
- [Towards Grounded Visual Spatial Reasoning in Multi-Modal Vision Language Models](https://mlsanity.com/paper/2308.09778/) (2023)
- [Unveiling the Compositional Ability Gap in Vision-Language Reasoning Model](https://mlsanity.com/paper/2505.19406/) (2025)
- [EasyARC: Evaluating Vision Language Models on True Visual Reasoning](https://mlsanity.com/paper/2506.11595/) (2025)
- [VISTA-Bench: Do Vision-Language Models Really Understand Visualized Text as Well as Pure Text?](https://mlsanity.com/paper/2602.04802/) (2026)
- [VRIQ: Benchmarking and Analyzing Visual-Reasoning IQ of VLMs](https://mlsanity.com/paper/2602.05382/) (2026)
- [Reasoning or Pattern Matching? Probing Large Vision-Language Models with Visual Puzzles](https://mlsanity.com/paper/2601.13705/) (2026)

## Built on by (most cited first, 6 of 6)

- [SeePhys Pro: Diagnosing Modality Transfer and Blind-Training Effects in Multimodal RLVR for Physics Reasoning](https://mlsanity.com/paper/2605.09266/) (2026)
- [MIRROR: Learning from the Other View for Multi-Modal Reasoning](https://mlsanity.com/paper/2607.21552/) (2026)
- [Which Source Wins? Task-Dependent Reliance in Vision-Language Models](https://mlsanity.com/paper/2608.17205/) (2026)
- [Transcribe, Then Reason: Two-Pass Decomposition for Multimodal Review](https://mlsanity.com/paper/2609.18958/) (2026)
- [Distilling Visual Reasoning into Text Space](https://mlsanity.com/paper/2609.34408/) (2026)
- [Learning How to Remember: Memory Modality Selection for GUI Agents](https://mlsanity.com/paper/2610.17928/) (2026)

## Abstract (arXiv)

Reasoning in vision-language models (VLMs) has recently attracted significant attention due to its broad applicability across diverse downstream tasks. However, it remains unclear whether the superior performance of VLMs stems from genuine vision-grounded reasoning or relies predominantly on the reasoning capabilities of their textual backbones. To systematically measure this, we introduce CrossMath, a novel multimodal reasoning benchmark designed for controlled cross-modal comparisons. Specifically, we construct each problem in text-only, image-only, and image+text formats guaranteeing identical task-relevant information, verified by human annotators. This rigorous alignment effectively isolates modality-specific reasoning differences while eliminating confounding factors such as information mismatch. Extensive evaluation of state-of-the-art VLMs reveals a consistent phenomenon: a substantial performance gap between textual and visual reasoning. Notably, VLMs excel with text-only inputs, whereas incorporating visual data (image+text) frequently degrades performance compared to the text-only baseline. These findings indicate that current VLMs conduct reasoning primarily in the textual space, with limited genuine reliance on visual evidence. To mitigate this limitation, we curate a CrossMath training set for VLM fine-tuning. Empirical evaluations demonstrate that fine-tuning on this training set significantly boosts reasoning performance across all individual and joint modalities, while yielding robust gains on two general visual reasoning tasks. Source code is available at https://github.com/xuyige/CrossMath.