trajectory-based action prediction on AgentNetBench
31.45Per-step Action ScoreFull Cache
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Full CacheModel=OpenCUA-7B, KV-cache budget=Full Cache2026.06 | 31.45 | 51.8 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 25.64 | 50.04 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 25.18 | 49.47 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 25.18 | 49.32 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 21.1 | 49.94 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=40%2026.06 | 20.67 | 47.31 | |
| Full CacheModel=UI-TARS-1.5-7B, KV-cache budget=Full Cache2026.06 | 20.34 | 49.75 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 20.02 | 48.92 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 18.1 | 47.01 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 18.04 | 45.3 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 18.04 | 41.17 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 17.44 | 47.22 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 16.45 | 47.31 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=20%2026.06 | 15.23 | 43.03 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 15.06 | 46.59 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 11.04 | 45.09 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=40%2026.06 | 10.3 | 45.45 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 10.26 | 42.67 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 9.39 | 39.16 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 9.14 | 39.65 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=10%2026.06 | 7.29 | 38.27 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 5 | 40.7 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 4.6 | 40.72 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 4.11 | 40.04 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=20%2026.06 | 3.52 | 39.46 | |
| STaR-KVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 3.42 | 35.89 | |
| GUIKVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 2.92 | 33.78 | |
| PyramidKVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 2.81 | 32.72 | |
| SnapKVModel=OpenCUA-7B, KV-cache budget=5%2026.06 | 2.81 | 33.62 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=10%2026.06 | 1.84 | 33.12 | |
| STaR-KVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 1.3 | 32.74 | |
| GUIKVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 1.25 | 32.63 | |
| SnapKVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 1.23 | 32.04 | |
| PyramidKVModel=UI-TARS-1.5-7B, KV-cache budget=5%2026.06 | 0.84 | 24.7 |