Share your thoughts, 1 month free Claude Pro on usSee more

Average Execution Time for Long-Horizon Composite Skills on OSExpert-Eval

18Execution Time (GIMP)

Human Expert

Updated 4mo ago

Evaluation Results

Method	Links
Human Expert 2026.03		18	25
OSExpert (Explore w/ Qwen-3-VL-8B) 2026.03		32	29
OSExpert (Exploration w/ GPT-5 & UI-TARS) 2026.03		34	27
OSExpert (Explore w/ Qwen-3-VL-8B, wo/ Fast Planner) 2026.03		46	31
OSExpert (Explore w/ Qwen-3-VL-8B, wo/ Boundary Check) 2026.03		84	91
Qwen-3-VL-8B 2026.03		139	136
OpenCUA-7B 2026.03		169	193
Computer-Use-Preview 2026.03		252	240
Agent-S3 w/ GPT-5 2026.03		998	1,231
CoAct-1 2026.03		1,260	1,104