Trajectory-based Action Prediction on AndroidControl
48.2Step AccuracyFull Cache
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Full CacheModel=UI-TARS-1.5-7B, KV-cache budget=Full Cache2026.06 | 48.2 | 49.75 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 48 | 49.94 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 46.4 | 48.92 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 45.2 | 45.45 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 43 | 46.59 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 43 | 47.31 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 42 | 47.22 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 41.8 | 45.09 | |
| Full CacheModel=OpenCUA-7B, KV-cache budget=Full Cache2026.06 | 37.8 | 51.8 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 36.8 | 49.32 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 36.8 | 50.04 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 36.4 | 40.7 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 35.8 | 49.47 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 35.2 | 40.72 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 35.2 | 47.31 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 35 | 40.04 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 34.4 | 47.01 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 34.2 | 39.46 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 33.2 | 41.17 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 31.8 | 45.3 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 29.6 | 38.27 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 29.2 | 42.67 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 28.8 | 43.03 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 25.4 | 33.12 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 24.2 | 39.65 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 23.4 | 39.16 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 21.8 | 32.63 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 21.6 | 32.74 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 21 | 32.04 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 19.8 | 24.7 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 19.8 | 33.62 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 17.6 | 35.89 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 17.4 | 33.78 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 13 | 32.72 |