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

TL;DR

MMIR introduces a dedicated benchmark to stress-test multimodal inconsistency reasoning in layout-rich artifacts by injecting five semantic error types across webpages, slides, and posters. It evaluates six MLLMs, finding proprietary models outperform open-source baselines and revealing gaps in cross-modal reasoning, especially with single-element inconsistencies and high layout density. The paper analyzes prompting strategies, finding CoT/SoM modest or inconsistent gains, but proposes Multimodal Interleaved CoT (MM-CoT) as a robust two-stage method that iteratively fuses visual and textual cues to improve detection and localization of inconsistencies. These findings highlight a critical bottleneck in current multimodal models and drive future work toward more integrated cross-modal reasoning and robust inconsistency detection in real-world content.

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.

Figures (7)

• Figure 1: An illustration of multimodal inconsistency reasoning on a webpage. An agent examines a webpage where the brand “IKEA AB” is mentioned, but other elements clearly refer to “Lorell.” Detecting this brand identity misattribution requires the ability to compare text fields across different sections of the page and reconcile them with accompanying images or context—an inherently multimodal reasoning task.
• Figure 2: There are five inconsistency categories in the MMIR benchmark, posing diverse challenges.
• Figure 3: MMIR Data filtering process.
• Figure 4: Fine-grained analysis of model performance.
• Figure 5: Model performance on layout complexity.