LLaDA-UI: Bringing Block-wise Diffusion to Vision-Language GUI Agents

Benchmark Model Rank Results
computer-use-agents-on-mobileworldLLaDA-UI#5Success Rate: 25.6
infographic-visual-question-answering-on-infovqaLLaDA-UI (pre-training)#6ANLS: 71.5
visual-question-answering-vqa-on-ai2d-test-with-maskLLaDA-UI (pre-training)#7Accuracy: 80.2
computer-use-agents-on-webvoyagerLLaDA-UI#9Success Rate: 56.9
visual-question-answering-vqa-on-simplevqaLLaDA-UI (pre-training)#11Accuracy: 44.3
computer-use-agents-on-osworld-verifiedLLaDA-UI#12Success Rate: 29.4
chart-question-answering-on-charxiv-dqLLaDA-UI (pre-training)#14Accuracy: 72.2
visual-question-answering-vqa-on-mmbench-cnLLaDA-UI (pre-training)#15Accuracy: 77.7
mathematical-reasoning-on-mathverseLLaDA-UI (pre-training)#18Accuracy: 48.0
object-counting-on-countbenchLLaDA-UI (pre-training)#20Accuracy: 86.2
visual-question-answering-vqa-on-mmbench-enLLaDA-UI (pre-training)#22Accuracy: 80.5
computer-use-agents-on-androidworldLLaDA-UI#25Success Rate: 53.5
chart-question-answering-on-charxiv-rqLLaDA-UI (pre-training)#28Accuracy: 35.6
computer-use-agents-on-screenspot-proLLaDA-UI#28Accuracy: 52.9
computer-use-agents-on-screenspot-v2LLaDA-UI#30Accuracy: 90.7
optical-character-recognition-ocr-on-ocrbenchLLaDA-UI (pre-training)#35Score: 855
visual-question-answering-on-docvqa-testLLaDA-UI (pre-training)#35ANLS: 0.915
chart-question-answering-on-chartqaLLaDA-UI (pre-training)#471:1 Accuracy: 84.8
multimodal-reasoning-on-mathvistaLLaDA-UI (pre-training)#64Accuracy: 70.5
multimodal-reasoning-on-mmmuLLaDA-UI (pre-training)#68Accuracy: 56.2
visual-question-answering-vqa-on-realworldqaLLaDA-UI (pre-training)#76Accuracy: 63.5
vision-language-understanding-on-mmstarLLaDA-UI (pre-training)#79Accuracy: 61.3