Mobile GUI automation on AndroidLab
61.6Success RateHiViG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HiViGPolicy Backbone=Gemini-3-Flash2026.06 | 61.6 | — | |
| Base agentPolicy Backbone=Gemini-3-Flash2026.06 | 58 | — | |
| OpenCUAPolicy Backbone=Gemini-3-Flash2026.06 | 57.2 | — | |
| SE-WSMPolicy Backbone=Gemini-3-Flash2026.06 | 57.2 | — | |
| CGI (32B)Policy Backbone=Gemini-3-Flash2026.06 | 55.8 | — | |
| UI-Venus-1.5-8BParams.=8B2026.02 | 55.1 | 68.1 | |
| CGI (8B)Policy Backbone=Gemini-3-Flash2026.06 | 55.1 | — | |
| GUI-Critic-R1Policy Backbone=Gemini-3-Flash2026.06 | 53.6 | — | |
| UI-Venus-1.5-30B-A3BParams.=30B2026.02 | 52.9 | 68.1 | |
| HiViGPolicy Backbone=Qwen3-VL-32B-Thinking2026.06 | 51.5 | — | |
| UI-Venus-1.0-72BParams.=72B2026.02 | 49.3 | — | |
| GUI-Critic-R1Policy Backbone=Qwen3-VL-32B-Thinking2026.06 | 49.3 | — | |
| Step-GUI-4BParams.=4B, Evaluated by=this paper2026.02 | 47.8 | — | |
| AutoGLM-MultilingualParams.=9B2026.02 | 47.7 | — | |
| OpenCUAPolicy Backbone=Qwen3-VL-32B-Thinking2026.06 | 47.1 | — | |
| CGI (32B)Policy Backbone=Qwen3-VL-32B-Thinking2026.06 | 46.9 | — | |
| AutoGLM-MobileParams.=9B2026.02 | 46.8 | — | |
| SE-WSMPolicy Backbone=Qwen3-VL-32B-Thinking2026.06 | 45.7 | — | |
| MobileUseParams.=72B2026.02 | 44.2 | — | |
| Base agentPolicy Backbone=Qwen3-VL-32B-Thinking2026.06 | 44.2 | — | |
| CGI (8B)Policy Backbone=Qwen3-VL-32B-Thinking2026.06 | 44.2 | — | |
| Qwen3-VL-8BParams.=8B, Evaluated by=this paper2026.02 | 43.5 | — | |
| Qwen3-VL-30B-A3BParams.=30B, Evaluated by=this paper2026.02 | 42 | — | |
| UI-Venus-1.0-7BParams.=7B2026.02 | 41.3 | — | |
| UI-GenieParams.=72B2026.02 | 41.2 | — | |
| V-DroidParams.=8B2026.02 | 38.3 | — | |
| UI-Venus-1.5-2BParams.=2B2026.02 | 36.2 | 44.2 | |
| T3A*Type=Fine-Tuned, Model=Qwen-2.5-7B-Instruct-ASL (Ours), Input Modality=Text2025.06 | 35.5 | — | |
| Qwen3-VL-2BParams.=2B, Evaluated by=this paper2026.02 | 33.3 | — | |
| T3A*Type=Fine-Tuned, Model=Qwen-2.5-7B-Instruct-SFT, Input Modality=Text2025.06 | 33.3 | — | |
| GPT-4o2026.02 | 31.2 | — | |
| AndroindLab-AgentType=Prompt-Driven, Model=GPT-4o, Input Modality=Image + Text2025.06 | 30.1 | — | |
| AndroindLab-AgentType=Prompt-Driven, Model=GPT-4o, Input Modality=Text2025.06 | 24.7 | — | |
| AndroindLab-AgentType=Prompt-Driven, Model=Claude-3.5-Sonnet, Input Modality=Image + Text2025.06 | 24.2 | — | |
| LLaMA3.1-ftParams.=8B2026.02 | 23.9 | — | |
| GLM4-9B-ftParams.=9B2026.02 | 21 | — | |
| AndroindLab-AgentType=Prompt-Driven, Model=Gemini-1.5-Pro, Input Modality=Text2025.06 | 17.2 | — | |
| Gemini-1.5-Pro2026.02 | 16.7 | — | |
| AndroindLab-AgentType=Prompt-Driven, Model=Gemini-1.5-Pro, Input Modality=Image + Text2025.06 | 13 | — |