GUI Grounding on OSWorld-G (test)
78.4Element AccuracyGTA1-32B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GTA1-32BModel Size=32B2025.12 | 78.4 | 63.2 | 73.3 | 65.2 | 0 | 65.2 | |
| MAI-UI-32BModel Size=32B, Zoom-In=true2025.12 | 75.2 | 78.5 | 78.3 | 62.4 | — | 70.9 | |
| MAI-UI-32BModel Size=32B, Zoom-In=false2025.12 | 72.4 | 73.6 | 73.9 | 57.7 | — | 67.6 | |
| UI-Venus-72BModel Size=72B2025.12 | 71.2 | 82.1 | 70.7 | 64.4 | — | 70.4 | |
| MAI-UI-8BModel Size=8B, Zoom-In=true2025.12 | 67.6 | 72.8 | 71.1 | 56.4 | — | 64.2 | |
| Seed1.5-VLProprietary=true2025.12 | 66.7 | 73.9 | 69.6 | 47 | 18.5 | 62.9 | |
| GTA1-7BModel Size=7B2025.12 | 65.7 | 42.1 | 62.7 | 56.1 | 0 | 55.1 | |
| GUI-Owl-32BModel Size=32B2025.12 | 64.5 | 67 | 67.2 | 45.6 | — | 58 | |
| GUI-Owl-7BModel Size=7B2025.12 | 63.6 | 64.8 | 61.3 | 41 | — | 55.9 | |
| Qwen3-VL-32B*Model Size=32B2025.12 | 63.3 | 72.8 | 66.4 | 51.7 | — | 60.6 | |
| MAI-UI-8BModel Size=8B, Zoom-In=false2025.12 | 63.3 | 72 | 66 | 51 | — | 60.1 | |
| UI-TARS-1.5-7BModel Size=7B2025.12 | 62.7 | 36.8 | 62.2 | 50.8 | 0 | 52.8 | |
| FOCUSUI-7Bvisual token retention ratio (r)=50%2026.01 | 62.1 | 64 | 63.6 | 31.5 | 0 | 54.1 | |
| FOCUSUI-7Bvisual token retention ratio (r)=100%2026.01 | 61.2 | 63.6 | 63.6 | 34.9 | 0 | 54.4 | |
| FOCUSUI-7Bvisual token retention ratio (r)=30%2026.01 | 60.9 | 63.6 | 64.4 | 31.5 | 0 | 53.9 | |
| UI-TARS-72BModel Size=72B2025.12 | 60.6 | 69.4 | 62.9 | 45.6 | 0 | 57.1 | |
| UI-Venus-7BModel Size=7B2025.12 | 60.5 | 74.6 | 61.5 | 45.5 | — | 58.8 | |
| FOCUSUI-3Bvisual token retention ratio (r)=50%2026.01 | 59.4 | 64.8 | 63.6 | 37.6 | 0 | 54.6 | |
| MAI-UI-2BModel Size=2B, Zoom-In=true2025.12 | 59.4 | 66.7 | 63.2 | 44.3 | — | 55.9 | |
| Trifuse+GUI-AIMAModel Size=7B, Training Protocol=Training-based (TB)2026.02 | 58.4 | — | — | — | — | — | |
| GUI-AIMAModel Size=3B, Training Protocol=Training-based (TB)2026.02 | 58.3 | — | — | — | — | — | |
| Trifuse+GUI-ActorModel Size=7B, Training Protocol=Training-based (TB)2026.02 | 57.9 | — | — | — | — | — | |
| UI-TARS-1.5-7BSize=7B2026.01 | 57.9 | 70.1 | 59.7 | 51.7 | 0 | 56 | |
| FOCUSUI-3Bvisual token retention ratio (r)=100%2026.01 | 57.6 | 65.9 | 59.7 | 37.6 | 0 | 53.4 | |
| FOCUSUI-3Bvisual token retention ratio (r)=30%2026.01 | 56.7 | 62.5 | 62.9 | 33.6 | 0 | 51.8 | |
| MAI-UI-2BModel Size=2B, Zoom-In=false2025.12 | 56.7 | 62.8 | 59.3 | 40.3 | — | 52 | |
| GUI-ActorModel Size=7B, Training Protocol=Training-based (TB)2026.02 | 56.6 | — | — | — | — | — | |
| Trifuse+GUI-ActorModel Size=3B, Training Protocol=Training-based (TB)2026.02 | 56.5 | — | — | — | — | — | |
| GUI-Actor-3BSize=3B2026.01 | 56.1 | 60.5 | 58.5 | 32.2 | 0 | 50.5 | |
| Jedi-7BSize=7B2026.01 | 55.5 | 65.9 | 57.7 | 46.9 | 7.4 | 54.1 | |
| Jedi-7BModel Size=7B2025.12 | 55.5 | 65.9 | 57.7 | 46.9 | 7.4 | 54.1 | |
| Qwen3-VL-8B*Model Size=8B2025.12 | 55.5 | 69 | 59.7 | 47.7 | — | 54.8 | |
| GUI-ActorModel Size=3B, Training Protocol=Training-based (TB)2026.02 | 54.6 | — | — | — | — | — | |
| GUI-Actor-7BSize=7B2026.01 | 54.2 | 60.2 | 58.1 | 30.9 | 0 | 49.5 | |
| Jedi-3BSize=3B2026.01 | 53 | 67.4 | 53.8 | 44.3 | 7.4 | 50.9 | |
| Jedi-3BModel Size=3B2025.12 | 53 | 67.4 | 53.8 | 44.3 | 7.4 | 50.9 | |
| UI-TARS-7BSize=7B2026.01 | 51.8 | 60.2 | 54.9 | 35.6 | 0 | 47.5 | |
| UI-TARS-7BModel Size=7B2025.12 | 51.8 | 60.2 | 54.9 | 35.6 | 0 | 47.5 | |
| Qwen3-VL-2B*Model Size=2B2025.12 | 45.8 | 61.7 | 54.2 | 39.6 | — | 45.9 | |
| Gemini-2.5-Pro2026.01 | 45.5 | 59.8 | 49 | 33.6 | 38.9 | 45.2 | |
| TrifuseModel Size=7B, Training Protocol=Training-free (TF)2026.02 | 43.6 | — | — | — | — | — | |
| Operator2026.01 | 42.4 | 51.3 | 46.6 | 31.5 | 0 | 40.6 | |
| OperatorProprietary=true2025.12 | 42.4 | 51.3 | 46.6 | 31.5 | 0 | 40.6 | |
| Aguvis-7BSize=7B2026.01 | 41.2 | 55.9 | 43.9 | 28.2 | 0 | 38.7 | |
| Aguvis-7BModel Size=7B2025.12 | 41.2 | 55.9 | 43.9 | 28.2 | 0 | 38.7 | |
| UGround-V1-7BSize=7B2026.01 | 40.3 | 51.3 | 43.5 | 24.8 | 0 | 36.4 | |
| UGround-7BModel Size=7B2025.12 | 40.3 | 51.3 | 43.5 | 24.8 | 0 | 36.4 | |
| TrifuseModel Size=3B, Training Protocol=Training-free (TF)2026.02 | 38.7 | — | — | — | — | — | |
| Qwen2.5-VL-7BSize=7B2026.01 | 32.7 | 45.6 | 41.9 | 18.1 | 0 | 31.4 | |
| OS-Atlas-7BModel Size=7B2025.12 | 29.4 | 44.1 | 35.2 | 16.8 | 7.4 | 27.7 | |
| Qwen2.5-VL-3BSize=3B2026.01 | 28.8 | 41.4 | 34.8 | 13.4 | 0 | 27.3 | |
| TAGModel Size=8.5B, Training Protocol=Training-free (TF)2026.02 | 25.3 | — | — | — | — | — | |
| OpenCUA-32BModel Size=32B2025.12 | — | — | — | — | — | 59.6 |