Dream-VL & Dream-VLA: Open Vision-Language and Vision-Language-Action Models with Diffusion Language Model Backbone

Benchmark Model Rank Results
visual-question-answering-vqa-on-seed-seedbenchDream-VL#11Accuracy: 76.4
mathematical-reasoning-on-mathverseDream-VL#30Accuracy: 31.5
robot-manipulation-on-simplerenv-widow-xDream-VLA#33Average: 0.604
chart-question-answering-on-chartqaDream-VL#491:1 Accuracy: 84.5
robot-manipulation-on-libero-longDream-VLA#49Success Rate: 95.0
robot-manipulation-on-libero-goalDream-VLA#51Success Rate: 97.2
video-question-answering-on-mlvuDream-VL#52Accuracy: 61.1
robot-manipulation-on-libero-spatialDream-VLA#60Success Rate: 97.6
multimodal-reasoning-on-mmmu-proDream-VL#61Accuracy: 26.0
robot-manipulation-on-libero-objectDream-VLA#64Success Rate: 98.8
visual-question-answering-vqa-on-realworldqaDream-VL#70Accuracy: 66.3
robot-manipulation-on-liberoDream-VLA#76Success Rate: 97.2
vision-language-understanding-on-mmstarDream-VL#84Accuracy: 59.9
multimodal-reasoning-on-mathvistaDream-VL#87Accuracy: 63.1
multimodal-reasoning-on-mmmuDream-VL#89Accuracy: 52.2