JMMMU-Pro: Image-based Japanese Multi-discipline Multimodal Understanding Benchmark via Vibe Benchmark Construction

TL;DR

JMMMU-Pro is a new Japanese multimodal benchmark created by combining images and text into a single visual question, challenging open-source models and advancing evaluation of Japanese multimodal understanding.

Contribution

The paper introduces JMMMU-Pro, a novel high-quality Japanese multimodal benchmark constructed via a scalable human-in-the-loop image generation process.

Findings

Open-source LMMs perform poorly on JMMMU-Pro

Vibe Benchmark Construction enables efficient benchmark creation

JMMMU-Pro provides a rigorous evaluation of Japanese multimodal understanding

Abstract

This paper introduces JMMMU-Pro, an image-based Japanese Multi-discipline Multimodal Understanding Benchmark, and Vibe Benchmark Construction, a scalable construction method. Following the evolution from MMMU to MMMU-Pro, JMMMU-Pro extends JMMMU by composing the question image and question text into a single image, thereby creating a benchmark that requires integrated visual-textual understanding through visual perception. To build JMMMU-Pro, we propose Vibe Benchmark Construction, a methodology in which an image generative model (e.g., Nano Banana Pro) produces candidate visual questions, and humans verify the outputs and, when necessary, regenerate with adjusted prompts to ensure quality. By leveraging Nano Banana Pro's highly realistic image generation capabilities and its ability to embed clean Japanese text, we construct a high-quality benchmark at low cost, covering a wide range of background and layout designs. Experimental results show that all open-source LMMs struggle substantially with JMMMU-Pro, underscoring JMMMU-Pro as an important benchmark for guiding future efforts in the open-source community. We believe that JMMMU-Pro provides a more rigorous evaluation tool for assessing the Japanese capabilities of LMMs and that our Vibe Benchmark Construction also offers an efficient guideline for future development of image-based VQA benchmarks.