Evaluator Accuracy on AndroidBench
94.9Overall AccuracyGUIDE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GUIDEBackbone=gemini-3.0-flash2026.04 | 94.9 | 85.6 | 98.1 | 100 | 95.8 | |
| Cap.+Mixtral2026.04 | 92.9 | — | — | — | — | |
| WebJudgeBackbone=gemini-3.0-flash2026.04 | 92.6 | 83.6 | 93.3 | 100 | 93.3 | |
| AgentTrekBackbone=gemini-3.0-flash2026.04 | 92.4 | 86.1 | 93.9 | 99.2 | 90.3 | |
| GPT-4V2026.04 | 90.6 | — | — | — | — | |
| Cap.+GPT-42026.04 | 89.8 | — | — | — | — | |
| QWen-VL2026.04 | 70.2 | — | — | — | — |