Desktop Task Success on macOSWorld
20.3Task Success RateRL + corrected reward estimator (rˆt)
Evaluation results
| Method | Links | |
|---|---|---|
| RL + corrected reward estimator (rˆt)Model Type=per-OS models2026.06 | 20.3 | |
| RL + raw evaluator reward (r˜t)Model Type=per-OS models2026.06 | 14.9 | |
| RL + corrected reward estimator (rˆt)Model Type=unified model2026.06 | 14.4 | |
| RL + raw evaluator reward (r˜t)Model Type=unified model2026.06 | 12.9 | |
| UI-TARSTraining Setup=Zero-shot baseline2026.06 | 8.4 |