Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models

TL;DR

This paper introduces MMIR, a benchmark to evaluate multimodal large language models' ability to detect and reason about semantic inconsistencies in complex, layout-rich visual-textual content, revealing current models' limitations.

Contribution

The paper presents MMIR, a new challenging benchmark with 534 samples for assessing MLLMs' inconsistency reasoning, and provides comprehensive evaluation and analysis of state-of-the-art models' performance.

Findings

Models with dedicated reasoning capabilities outperform others.

Open-source models are particularly vulnerable to inconsistencies.

Single-modality prompting yields marginal improvements.

Abstract

Existing Multimodal Large Language Models (MLLMs) are predominantly trained and tested on consistent visual-textual inputs, leaving open the question of whether they can handle inconsistencies in real-world, layout-rich content. To bridge this gap, we propose the Multimodal Inconsistency Reasoning (MMIR) benchmark to assess MLLMs' ability to detect and reason about semantic mismatches in artifacts such as webpages, presentation slides, and posters. MMIR comprises 534 challenging samples, each containing synthetically injected errors across five reasoning-heavy categories: Factual Contradiction, Identity Misattribution, Contextual Mismatch, Quantitative Discrepancy, and Temporal/Spatial Incoherence. We evaluate six state-of-the-art MLLMs, showing that models with dedicated multimodal reasoning capabilities, such as o1, substantially outperform their counterparts while open-source models remain particularly vulnerable to inconsistency errors. Detailed error analyses further show that models excel in detecting pairwise inconsistencies but struggle with inconsistencies confined to single elements in complex layouts. Probing experiments reveal that single-modality prompting, including Chain-of-Thought (CoT) and Set-of-Mark (SoM) methods, yields marginal gains, revealing a key bottleneck in cross-modal reasoning. Our findings highlight the need for advanced multimodal reasoning and point to future research on multimodal inconsistency.