MME-CoT: Benchmarking Chain-of-Thought in Large Multimodal Models for Reasoning Quality, Robustness, and Efficiency

Dongzhi Jiang, Renrui Zhang, Ziyu Guo, Yanwei Li, Yu Qi, Xinyan Chen, Liuhui Wang, Jianhan Jin, Claire Guo, Shen Yan, Bo Zhang, Chaoyou Fu, Peng Gao, Hongsheng Li

TL;DR: A Comprehensive Evaluation Suite for Multimodal Reasoning

Abstract:

Answering questions with Chain-of-Thought (CoT) has significantly enhanced the reasoning capabilities of Large Language Models (LLMs), yet its impact on Large Multimodal Models (LMMs) still lacks a systematic assessment and in-depth investigation. In this paper, we introduce **MME-CoT**, a specialized benchmark evaluating the CoT reasoning performance of LMMs, spanning six domains: math, science, OCR, logic, space-time, and general scenes.
As the first comprehensive study in this area, we propose a thorough evaluation suite incorporating three novel metrics that assess the reasoning quality, robustness, and efficiency at a fine-grained level.
Leveraging curated high-quality data and a unique evaluation strategy, we conduct an in-depth analysis of state-of-the-art LMMs, uncovering several key insights: *1)* Models with reflection mechanism demonstrate a superior CoT quality, with Kimi k1.5 outperforming GPT-4o and demonstrating the highest quality results; *2)* CoT prompting often degrades LMM performance on perception-heavy tasks, suggesting a potentially harmful overthinking behavior; and *3)* Although the CoT quality is high, LMMs with reflection exhibit significant inefficiency in both normal response and self-correction phases.
We hope MME-CoT serves as a foundation for advancing multimodal reasoning in LMMs.

Lay Summary: This paper introduces MME-CoT, a new benchmark for testing whether AI systems that process both text and images can reason through problems step-by-step, rather than just checking final answers. Testing 16 AI models across math, science, and visual reasoning tasks, the researchers found that while self-correcting models generally perform better, step-by-step reasoning often hurts performance on simple visual tasks (AI "overthinks" easy problems) and about 40% of self-correction attempts don't actually help solve problems, revealing important flaws in how current AI systems think.

Link To Code: https://github.com/MME-Benchmarks/MME-CoT

Primary Area: Deep Learning->Everything Else

Keywords: Evaluation and Benchmark, MLLM, Reasoning

Submission Number: 1180