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

GUI Agent Interaction on OSWorld w/o Loop

29.5AUV

GPT-4o + ScaleCUA-7B

Updated 1mo ago

Evaluation Results

Method	Links
GPT-4o + ScaleCUA-7B 2026.02		29.5
UI-TARS-1.5-7B 2026.02		28
UI-TARS-72B-DPO 2026.02		26.3
Claude3.7-Sonnet 2026.02		9
Qwen2.5-VL-72B-Instruct 2026.02		8.3