Operating System Agent Control on WindowsAgentArena
21.7Success RateUltraCUA-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| UltraCUA-7B2025.10 | 21.7 | |
| UI-TARS-1.5-7B2025.10 | 18.1 | |
| Qwen2-VL-7BTraining Data=OpenCUA Data2025.10 | 13.5 | |
| OS-SYMPHONYBackbone=GPT-5, Step=502026.01 | 0.635 | |
| OS-SYMPHONYBackbone=GPT-5-Mini, Step=502026.01 | 0.622 | |
| Agent S3Backbone=GPT-5, Step=1002026.01 | 0.566 | |
| Agent S3Backbone=GPT-5, Step=502026.01 | 0.541 | |
| UI-TARS-2Step=502026.01 | 0.506 | |
| OS-SYMPHONYBackbone=Qwen3-VL-32B-Instruct, Step=502026.01 | 0.453 | |
| UI-TARS-1.5-7BStep=502026.01 | 0.421 | |
| Qwen3-VL-32B-InstructStep=502026.01 | 0.317 |