GUI agent task on OSWORLD noisy (Overall)
29Pass@8 Success RateUI-TARS-1.5-7B + ENVS
Evaluation Results
| Method | Links | |
|---|---|---|
| UI-TARS-1.5-7B + ENVSModel=UI-TARS-1.5-7B + ENVS, Train data=noisy trajectories2026.06 | 29 | |
| UI-TARS-1.5-7B + ARPOModel=UI-TARS-1.5-7B + ARPO, Train data=noisy rollouts2026.06 | 21.7 | |
| UI-TARS-1.5-7BModel=UI-TARS-1.5-7B, Train data=—2026.06 | 20.3 |