InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing

Open paper
Benchmark Model Rank Results
image-editing-on-risebenchInternVL-U (4B, w/ CoT)#9RISEBench Overall Accuracy: 9.4
text-to-image-generation-on-longtext-bench-zhInternVL-U (4B)#14Accuracy: 86.0
text-to-image-generation-on-oneig-zhInternVL-U (4B)#15OneIG Overall Score: 0.5
text-to-image-generation-on-cvtg-2kInternVL-U (4B)#19Word Accuracy: 0.623
text-to-image-generation-on-oneig-enInternVL-U (4B)#22OneIG Overall Score: 0.5
mathematical-reasoning-on-mathverseInternVL-U (4B)#23Accuracy: 45.6
image-editing-on-gedit-bench-enInternVL-U (4B, w/ CoT)#24Overall: 6.88
image-generation-on-wiseInternVL-U (4B, w/ CoT)#24Overall: 0.58
text-to-image-generation-on-dpg-benchInternVL-U (4B)#58DPG-Bench Score: 85.18
text-to-image-generation-on-genevalInternVL-U (4B)#62Overall: 0.85
multimodal-reasoning-on-mmmuInternVL-U (4B)#76Accuracy: 54.7
chart-question-answering-on-chartqaInternVL-U (4B)#771:1 Accuracy: 76.6

Rank counts only results with a code link.