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

Fine-Grained Action Execution on OSExpert-Eval (Execution Time)

22GIMP Execution Time (s)

Human Expert

Updated 4mo ago

Evaluation Results

Method	Links
Human Expert 2026.03		22	11
OSExpert (Explore w/ Qwen-3-VL-8B) 2026.03		35	39
OSExpert (Exploration w/ GPT-5 & UI-TARS) 2026.03		35	37
OSExpert (Explore w/ Qwen-3-VL-8B, wo/ Fast Planner) 2026.03		38	44
OSExpert (Explore w/ Qwen-3-VL-8B, wo/ Boundary Check) 2026.03		61	82
Qwen-3-VL-8B 2026.03		83	138
OpenCUA-7B 2026.03		162	173
Computer-Use-Preview 2026.03		256	282
Agent-S3 w/ GPT-5 2026.03		702	633
CoAct-1 2026.03		911	782