Video2GUI: Synthesizing Large-Scale Interaction Trajectories for Generalized GUI Agent Pretraining

Benchmark Model Rank Results
computer-use-agents-on-osw-g-osworld-gMimo-VL-7B + WildGUI#7Accuracy: 67.6
computer-use-agents-on-screenspot-proMimo-VL-7B + WildGUI#25Accuracy: 56.9
computer-use-agents-on-androidworldMimo-VL-7B + WildGUI#30Success Rate: 31.9