List your company

Web Navigation and Task Automation on VisualWebArena 910 tasks (Full)

88.7Success Rate (%)Human
8.30829.17950.0570.921May 24, 2026

Evaluation results

MethodLinks
Human2026.0588.7———
PANDOOptimization=Skill-library and supplementary optimization2026.0558.30.0850.0851
SGV (Gemini-2.5 Flash)Backbone=Gemini-2.5 Flash, Verification=Two-pass self-grounded verifier2026.05540.3710.3712.2
WALTBudget=One-time offline tool-discovery included in amortized2026.0545.20.5920.641—
GPT-5.2 (M) + SoM (Qwen-2.5VL)Modalities=Multimodal (M), Prompting=SoM, Vision Model=Qwen-2.5VL2026.0538.40.2520.2520.77
GPT-5.2 (M) + SoM (BLIP-2)Modalities=Multimodal (M), Prompting=SoM, Vision Model=BLIP-22026.0533.20.2580.2580.79
GPT-5.2 + SoM (Qwen-2.5VL)Prompting=SoM, Vision Model=Qwen-2.5VL2026.0531.60.3180.3180.97
GPT-5.2 + CaptionPrompting=Caption2026.0524.80.3450.3451.05
GPT-5.2 Text-OnlyPrompting=Text-Only2026.0511.40.3280.3281