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

GUI Agent Task on OSWORLD clean (held-out)

7.5Pass@8 Success Rate

UI-TARS-1.5-7B + ENVS

Updated 1mo ago

Evaluation Results

Method	Links
UI-TARS-1.5-7B + ENVS 2026.06		7.5
UI-TARS-1.5-7B + ARPO 2026.06		4.7
UI-TARS-1.5-7B 2026.06		2.3