MMMG: A Comprehensive and Reliable Benchmark for Multitask Multimodal Generation
Aug 1, 2026·,,,,,,
,,,,,·
1 min read
Jihan Yao
Yushi Hu
Wenyuan Wang
Bin Han
Guang Yang
Yujie Yi
Shangbin Feng
Bingbing Wen
Ranjay Krishna
Lucy Lu Wang
Yulia Tsvetkov
Noah A. Smith
Banghua Zhu
Abstract
We present MMMG, a comprehensive and reliable evaluation suite for multitask multimodal generation that covers diverse tasks and modalities to better assess the capabilities of large multimodal models.
Type
Publication
COLM 2026
MMMG is an evaluation suite for multitask multimodal generation, enabling more reliable assessment of large multimodal models across a wide range of generation tasks and modalities.
