Mobile UI Task Automation on AndroidWorld (full)
82.79Success Rate (Easy)Gemini-3-Flash
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Gemini-3-FlashAgent Framework=M3A, World-model Feedback=Full Text2026.05 | 82.79 | 61.81 | 31.58 | 67.89 | 61,207.18 | 12.17 | 1,042.4 | |
| Gemini-3-FlashAgent Framework=M3A, World-model Feedback=Delta Text2026.05 | 81.15 | 58.33 | 35.53 | 66.59 | 56,767.73 | 11.69 | 980.39 | |
| GPT-5.4Agent Framework=M3A, World-model Feedback=Delta Text2026.05 | 76.23 | 59.72 | 31.58 | 63.79 | 48,557 | 12.46 | 1,017.37 | |
| GPT-5.4Agent Framework=M3A, World-model Feedback=None (Baseline)2026.05 | 72.68 | 41.67 | 29.82 | 56.03 | 2,178.89 | 11.44 | 220.6 | |
| Gemini-3-FlashAgent Framework=M3A, World-model Feedback=Code2Image2026.05 | 72.13 | 44.44 | 10.53 | 53.44 | 217,123.29 | 12.72 | 2,714.06 | |
| GPT-5.4Agent Framework=M3A, World-model Feedback=Full Text2026.05 | 72.13 | 54.17 | 34.21 | 60.34 | 56,754.34 | 12.61 | 1,353.21 | |
| GPT-5.4Agent Framework=M3A, World-model Feedback=Code2Image2026.05 | 69.67 | 56.94 | 26.32 | 58.62 | 224,245.86 | 12.31 | 3,107.67 | |
| Qwen3-VL-32BAgent Framework=M3A, World-model Feedback=Delta Text2026.05 | 64.75 | 50.69 | 21.05 | 53.23 | 47,497.21 | 13.87 | 958.14 | |
| Qwen3-VL-32BAgent Framework=M3A, World-model Feedback=Full Text2026.05 | 63.11 | 50 | 21.05 | 52.16 | 51,497.53 | 14.1 | 1,095.01 | |
| Qwen3-VL-32BAgent Framework=M3A, World-model Feedback=Code2Image2026.05 | 63.11 | 43.06 | 18.42 | 49.57 | 209,073.95 | 13.89 | 2,490.72 | |
| Qwen3-VL-32BAgent Framework=M3A, World-model Feedback=None (Baseline)2026.05 | 62.57 | 44.9 | 24.56 | 50.86 | 1,907.01 | 13.78 | 553.32 | |
| Gemini-3-FlashAgent Framework=M3A, World-model Feedback=None (Baseline)2026.05 | 62.29 | 44.44 | 24.56 | 50.57 | 1,238.86 | 11.05 | 290.61 | |
| Qwen3-VL-8BAgent Framework=M3A, World-model Feedback=Delta Text2026.05 | 59.01 | 34.72 | 17.1 | 44.61 | 54,461.79 | 13.02 | 439.93 | |
| Qwen3-VL-8BAgent Framework=M3A, World-model Feedback=Code2Image2026.05 | 54.92 | 31.94 | 13.16 | 40.95 | 214,360.66 | 13.44 | 2,600.86 | |
| Qwen3-VL-8BAgent Framework=M3A, World-model Feedback=None (Baseline)2026.05 | 54.64 | 27.77 | 16.66 | 40.08 | 1,698.13 | 13.69 | 144.56 | |
| Qwen3-VL-8BAgent Framework=M3A, World-model Feedback=Full Text2026.05 | 50.82 | 33.33 | 11.84 | 39.01 | 58,548.32 | 13.87 | 1,469.09 |