Grounding Computer Use Agents on Human Demonstrations

Benchmark Model Rank Results
computer-use-agents-on-ui-v-ui-visionGroundNext-3B (RL)#2Accuracy: 62.1
computer-use-agents-on-osw-g-osworld-gGroundNext-7B (RL)#5Accuracy: 67.7
computer-use-agents-on-mmb-gui-mmbench-guiGroundNext-7B (RL)#9Accuracy: 81.1
computer-use-agents-on-screenspot-proGroundNext-7B (RL)#27Accuracy: 52.9
computer-use-agents-on-screenspot-v2GroundNext-7B (RL)#33Accuracy: 90.4