Uni-MMMU: A Massive Multi-discipline Multimodal Unified Benchmark

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zou, Kai, Huang, Ziqi, Dong, Yuhao, Tian, Shulin, Zheng, Dian, Liu, Hongbo, He, Jingwen, Liu, Bin, Qiao, Yu, Liu, Ziwei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911602303827968
author Zou, Kai
Huang, Ziqi
Dong, Yuhao
Tian, Shulin
Zheng, Dian
Liu, Hongbo
He, Jingwen
Liu, Bin
Qiao, Yu
Liu, Ziwei
author_facet Zou, Kai
Huang, Ziqi
Dong, Yuhao
Tian, Shulin
Zheng, Dian
Liu, Hongbo
He, Jingwen
Liu, Bin
Qiao, Yu
Liu, Ziwei
contents Unified multimodal models aim to jointly enable visual understanding and generation, yet current benchmarks rarely examine their true integration. Existing evaluations either treat the two abilities in isolation or overlook tasks that inherently couple them. To address this gap, we present Uni-MMMU, a comprehensive and discipline-aware benchmark that systematically unfolds the bidirectional synergy between generation and understanding across eight reasoning-centric domains, including science, coding, mathematics, and puzzles. Each task is bidirectionally coupled, demanding models to (i) leverage conceptual understanding to guide precise visual synthesis, or (ii) utilize generation as a cognitive scaffold for analytical reasoning. Uni-MMMU incorporates verifiable intermediate reasoning steps, unique ground truths, and a reproducible scoring protocol for both textual and visual outputs. Through extensive evaluation of state-of-the-art unified, generation-only, and understanding-only models, we reveal substantial performance disparities and cross-modal dependencies, offering new insights into when and how these abilities reinforce one another, and establishing a reliable foundation for advancing unified models.
format Preprint
id arxiv_https___arxiv_org_abs_2510_13759
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Uni-MMMU: A Massive Multi-discipline Multimodal Unified Benchmark
Zou, Kai
Huang, Ziqi
Dong, Yuhao
Tian, Shulin
Zheng, Dian
Liu, Hongbo
He, Jingwen
Liu, Bin
Qiao, Yu
Liu, Ziwei
Computer Vision and Pattern Recognition
Unified multimodal models aim to jointly enable visual understanding and generation, yet current benchmarks rarely examine their true integration. Existing evaluations either treat the two abilities in isolation or overlook tasks that inherently couple them. To address this gap, we present Uni-MMMU, a comprehensive and discipline-aware benchmark that systematically unfolds the bidirectional synergy between generation and understanding across eight reasoning-centric domains, including science, coding, mathematics, and puzzles. Each task is bidirectionally coupled, demanding models to (i) leverage conceptual understanding to guide precise visual synthesis, or (ii) utilize generation as a cognitive scaffold for analytical reasoning. Uni-MMMU incorporates verifiable intermediate reasoning steps, unique ground truths, and a reproducible scoring protocol for both textual and visual outputs. Through extensive evaluation of state-of-the-art unified, generation-only, and understanding-only models, we reveal substantial performance disparities and cross-modal dependencies, offering new insights into when and how these abilities reinforce one another, and establishing a reliable foundation for advancing unified models.
title Uni-MMMU: A Massive Multi-discipline Multimodal Unified Benchmark
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.13759