MathSight: A Benchmark Exploring Have Vision-Language Models Really Seen in University-Level Mathematical Reasoning?

TL;DR

MathSight is a benchmark that evaluates how much visual information actually contributes to multimodal mathematical reasoning, revealing that visual input's role diminishes with difficulty and challenging assumptions about VLM capabilities.

Contribution

The paper introduces MathSight, a novel benchmark with varied visual problem variants to isolate and measure the true impact of visual information in vision-language models.

Findings

Visual contribution decreases as problem difficulty increases.

Qwen3-VL without images outperforms multimodal variants and GPT-5.

Benchmarks like MathSight are essential for advancing genuine visual reasoning.

Abstract

Recent advances in Vision-Language Models (VLMs) have achieved impressive progress in multimodal mathematical reasoning. Yet, how much visual information truly contributes to reasoning remains unclear. Existing benchmarks report strong overall performance but seldom isolate the role of the image modality, leaving open whether VLMs genuinely leverage visual understanding or merely depend on linguistic priors. To address this, we present MathSight, a university-level multimodal mathematical reasoning benchmark designed to disentangle and quantify the effect of visual input. Each problem includes multiple visual variants -- original, hand-drawn, photo-captured -- and a text-only condition for controlled comparison. Experiments on state-of-the-art VLMs reveal a consistent trend: the contribution of visual information diminishes with increasing problem difficulty. Remarkably, Qwen3-VL without any image input surpasses both its multimodal variants and GPT-5, underscoring the need for benchmarks like MathSight to advance genuine vision-grounded reasoning in future models.