List your company
ResearchFormerly WizWand

GUI Grounding on ScreenSpot v2 (test)

95.6Element AccuracyVISTA
49.42461.41273.485.388Feb 6, 2026Feb 27, 2026Mar 20, 2026Apr 10, 2026May 1, 2026May 22, 2026Jun 12, 2026
Updated 4mo ago

Evaluation results

MethodLinks
VISTAModel Scale=≈8B, Inference Method=Multi-view (+ MVP)2026.0695.6
VISTAModel Scale=≈8B, Inference Method=Single-view2026.0695.5
VISTAModel Scale=≥30B, Inference Method=Multi-view (+ MVP)2026.0695.4
UI-VenusModel Scale=72B2026.0695.3
MAI-UIModel Scale=≈8B2026.0695.2
GTA1Model Scale=≥30B2026.0695.2
VISTAModel Scale=≥30B, Inference Method=Single-view2026.0695.2
Step-GUIModel Scale=≈8B2026.0695.1
Holo2Model Scale=≥30B2026.0694.9
Qwen3-VLModel Scale=≥30B2026.0694.7
VISTAModel Scale=≈4B, Inference Method=Multi-view (+ MVP)2026.0694.5
VISTAModel Scale=≈4B, Inference Method=Single-view2026.0694.4
UI-VenusModel Scale=≈8B2026.0694.1
Step-GUIModel Scale=≈4B2026.0693.6
OpenCUAModel Scale=≥30B2026.0693.4
Trifuse+GUI-ActorModel Size=7B, Training Protocol=Training-based (TB)2026.0293.2
Holo2Model Scale=≈8B2026.0693.2
GUI-OwlModel Scale=≥30B2026.0693.1
Trifuse+GUI-AIMAModel Size=7B, Training Protocol=Training-based (TB)2026.0292.9
OpenCUAModel Scale=72B2026.0692.9
GUI-OwlModel Scale=≈8B2026.0692.8
GTA1Model Scale=≈8B2026.0692.4
OpenCUAModel Scale=≈8B2026.0692.3
GUI-ActorModel Size=7B, Training Protocol=Training-based (TB)2026.0292.1
Qwen3-VLModel Scale=≈8B2026.0691.7
UI-TARS-1.5Model Scale=≈8B2026.0691.6
Trifuse+GUI-ActorModel Size=3B, Training Protocol=Training-based (TB)2026.0291.5
GUI-AIMAModel Size=3B, Training Protocol=Training-based (TB)2026.0291.5
GUI-ActorModel Size=3B, Training Protocol=Training-based (TB)2026.0291
Qwen3-VLModel Scale=≈4B2026.0688.5
TrifuseModel Size=7B, Training Protocol=Training-free (TF)2026.0286.9
TrifuseModel Size=3B, Training Protocol=Training-free (TF)2026.0282.6
TAGModel Size=8.5B, Training Protocol=Training-free (TF)2026.0251.2