List your company
ResearchFormerly WizWand

Multimodal Image Understanding on MMMU

54.2ScoreBAGEL-Uni-Edit (Ours)
40.99244.42147.8551.279Oct 27, 2025Nov 30, 2025Jan 3, 2026Feb 6, 2026Mar 12, 2026Apr 15, 2026May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BAGEL-Uni-Edit (Ours)Training Stage=Stage 22026.0554.2
MRoPE-IBackbone=Qwen3-VL-8B-Instruct2025.1053.89
BAGEL-Uni-Edit (Ours)Training Stage=Stage 12026.0553.6
MHRoPEBackbone=Qwen3-VL-8B-Instruct2025.1053.33
Uni-Edit w/o VAE dropoutVAE dropout=false2026.0552.9
BAGEL2026.0552.8
BAGEL-RecA2026.0552.8
Uni-Edit w/o joint trainingjoint training=false2026.0552.8
Uni-Edit ViT (und.) 224x224Resolution=224x2242026.0552.5
BAGEL-AnyEdit2026.0552.2
HoPEBackbone=Qwen3-VL-8B-Instruct2025.1051.89
VideoRoPEBackbone=Qwen3-VL-8B-Instruct2025.1051.78
Uni-Edit ViT (und.) 224x980Resolution=224x9802026.0551.7
Vanilla RoPEBackbone=Qwen3-VL-8B-Instruct2025.1051.18
Uni-Edit ViT (und.) 378x980Resolution=378x9802026.0551.1
CircleRoPEBackbone=Qwen3-VL-8B-Instruct2025.1050.67
MRoPEBackbone=Qwen3-VL-8B-Instruct2025.1050.11
MRoPE-IBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE-I2025.1046.33
MHRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MHRoPE2025.1046.11
Vanilla RoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=Vanilla RoPE2025.1045.78
HoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=HoPE2025.1045.44
CircleRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=CircleRoPE2025.1045.33
VideoRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=VideoRoPE2025.1044.44
+Uni-Edit (Ours)Base Model=Janus-Pro2026.0543.5
MRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE2025.1042.89
Janus-Pro2026.0541.5