OS-level Agent Task Execution on OSWorld subset
46.48Success RateEVA-CUA-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EVA-CUA-8BDeployment Mode=teacher-free test-time policy2026.06 | 46.48 | — | — | |
| SFT collector with SRCDeployment Mode=teacher-assisted data collection2026.06 | 43.27 | 16.82 | 5.29 | |
| Expert SFT + SRCDeployment Mode=teacher-free test-time policy2026.06 | 40.15 | 11.86 | — | |
| Expert SFTDeployment Mode=teacher-free test-time policy2026.06 | 27.22 | 12.81 | — | |
| Base modelDeployment Mode=teacher-free test-time policy2026.06 | 23.47 | 12.88 | — |