GUI Planning on AndroidControl Low
95.05SR (%)UI-TARS-72B-SFT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| UI-TARS-72B-SFTOnline Data=Y, Training Steps=–, Training Trajectories=145K2026.04 | 95.05 | — | 98.17 | |
| UI-TARS-7B-SFTOnline Data=Y, Training Steps=–, Training Trajectories=145K2026.04 | 94.81 | — | 98.08 | |
| AgentCPM-GUI-8BOnline Data=N, Training Steps=>470K, Training Trajectories=>55K2026.04 | 88.6 | — | 92.8 | |
| SOLAR-RLOnline Data=N, Training Steps=94K, Training Trajectories=15K2026.04 | 88.57 | — | 93.24 | |
| Doubao + ICMBase Model=Doubao, Refinement=ICM2026.01 | 86.4 | 86.6 | — | |
| GUI-Owl-7BOnline Data=Y, Training Steps=–, Training Trajectories=–2026.04 | 86.25 | — | 91.05 | |
| DoubaoMethod Type=N/A, Size=N/A, Note=Closed source UI-TARS 1.52026.01 | 86.2 | 86.4 | — | |
| UI-Venus-Navi-7BOnline Data=N, Training Steps=350K, Training Trajectories=–2026.04 | 86.16 | — | 92.17 | |
| Doubao + ICM-r2Base Model=Doubao, Refinement=ICM-r22026.01 | 86 | 86.6 | — | |
| Qwen2.5-VL-7BOnline Data=–, Training Steps=–, Training Trajectories=–2026.04 | 85.05 | — | 94.61 | |
| Qwen 2.5 VLMethod Type=N/A, Size=7B2026.01 | 81.8 | 85.6 | — | |
| Qwen 2.5 VL + ICM-r2Base Model=Qwen 2.5 VL, Refinement=ICM-r22026.01 | 81.8 | 85.4 | — | |
| GLM-4.5V (106B)Online Data=–, Training Steps=–, Training Trajectories=–2026.04 | 81.37 | — | 86.35 | |
| Qwen 2.5 VL + ICMBase Model=Qwen 2.5 VL, Refinement=ICM2026.01 | 81.2 | 86 | — | |
| GLM-4.1V-ThinkingOnline Data=–, Training Steps=–, Training Trajectories=–2026.04 | 80.66 | — | 86.09 | |
| AguvisMethod Type=SFT, Size=7B2026.01 | 80.5 | — | — | |
| UI-TARS 1.0* + ICMBase Model=UI-TARS 1.0, Refinement=ICM2026.01 | 80.4 | 87.6 | — | |
| UI-TARS 1.0* + ICM-r2Base Model=UI-TARS 1.0, Refinement=ICM-r22026.01 | 80.2 | 86.8 | — | |
| UI-TARS 1.5* + ICM-r2Base Model=UI-TARS 1.5, Refinement=ICM-r22026.01 | 79.2 | 85.5 | — | |
| UI-TARS 1.5* + ICMBase Model=UI-TARS 1.5, Refinement=ICM2026.01 | 79 | 85 | — | |
| UI-TARS 1.0*Method Type=N/A, Size=7B, Note=Reproduction2026.01 | 75.4 | 85.1 | — | |
| SeeClickMethod Type=SFT, Size=9.6B2026.01 | 75 | 73.4 | — | |
| WebChain-LCRL-7B +SGRL+CoT-SFTTraining Paradigm=Trained on WebChain2026.03 | 74.1 | 84.8 | 86.2 | |
| UI-TARS 1.5*Method Type=N/A, Size=7B, Note=Reproduction2026.01 | 72.2 | 82.4 | — | |
| Qwen2.5-VL-7BTraining Paradigm=Zero Shot2026.03 | 72 | 83 | 87 | |
| Aria-UIMethod Type=SFT, Size=3.9B2026.01 | 67.3 | 87.7 | — | |
| WebChain-LCRL-3B +SGRL+CoT-SFTTraining Paradigm=Trained on WebChain2026.03 | 67.2 | 83.1 | 85.2 | |
| GUI-R1-7BTraining Paradigm=Trained on Other Datasets2026.03 | 66.5 | 84 | 85.2 | |
| UI-R1Method Type=RFT, Size=3B2026.01 | 66.4 | 82.4 | — | |
| GUI-R1Method Type=RFT, Size=3B2026.01 | 64.4 | 81.6 | — | |
| GUI-R1-3BTraining Paradigm=Trained on Other Datasets2026.03 | 64.4 | 81.6 | 83.7 | |
| WebChain-LCRL-3B +CoT-SFTTraining Paradigm=Trained on WebChain2026.03 | 64.4 | 79.1 | 80.1 | |
| WebChain-LCRL-7B +CoT-SFTTraining Paradigm=Trained on WebChain2026.03 | 63.7 | 77.3 | 81.1 | |
| WebChain-LCRL-7BTraining Paradigm=Trained on WebChain2026.03 | 62.6 | 75.5 | 79 | |
| WebChain-LCRL-3BTraining Paradigm=Trained on WebChain2026.03 | 60.8 | 76.2 | 77.9 | |
| Qwen2.5-VL-3BTraining Paradigm=Zero Shot2026.03 | 60 | 74.6 | 61 | |
| OS-Atlas-BaseMethod Type=ZS, Size=7B2026.01 | 50.9 | 73.4 | — | |
| GPT4o + ICMBase Model=GPT4o, Refinement=ICM2026.01 | 24.8 | 9.2 | — | |
| GPT4o + ICM-r2Base Model=GPT4o, Refinement=ICM-r22026.01 | 23.8 | 8.5 | — | |
| GPT4oMethod Type=N/A, Size=N/A2026.01 | 20.4 | 8 | — |