List your company

Action Success Prediction on AndroidWorld (smallest held-out split)

87.1PrecisionVisCritic
72.64476.39780.1583.903Jun 23, 2026

Evaluation results

MethodLinks
VisCriticBase Model=Qwen2.5-VL base, Modality=Visual2026.0687.183.485.2
BacktrackAgentBase Model=Qwen2.5-VL base, Modality=Text2026.0675.664.369.5
GUI-PRABase Model=Qwen2.5-VL base, Modality=Text2026.0673.267.870.4