GUI Agent Task Execution on AndroidWorld, MobileMiniWob++, and DroidTask
2.1Latency (s)AutoDroid-V2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AutoDroid-V2Type=Llama-3-8B-ft, Input=SoM2026.05 | 2.1 | — | |
| EAMType=GPT-4o, Qwen2.5-3B-instruct-ft, Input=SoM2026.05 | 2.8 | 8.3 | |
| UI-TARS-2BType=UI-TARS-2B, Input=screen2026.05 | 6 | — | |
| Qwen2.5-VL-3BType=Qwen 2.5-VL-3B, Input=SoM2026.05 | 7.7 | — | |
| UI-TARS-7BType=UI-TARS-7B, Input=screen2026.05 | 8.8 | 32.7 | |
| GPT-4oType=GPT-4o, Input=SoM2026.05 | 9.3 | 50.8 | |
| AppAgentXType=GPT-4o, Input=SoM2026.05 | 16 | 6.2 | |
| M3AType=GPT-4o, Input=SoM2026.05 | 16.9 | 62.6 | |
| GUI-ExplorerType=GPT-4o, Input=SoM2026.05 | 66.4 | 73.1 |