List your company
ResearchFormerly WizWand

E-commerce Risk Management Multi-step Trajectory Evaluation on RISK-Bench Multi-step

82.8Multi-step Success RateRISK-R1-7B
-3.31219.04441.463.756Sep 26, 2025
Updated 6mo ago

Evaluation results

MethodLinks
RISK-R1-7BModel Category=Ours, Base Model=Qwen2.5-VL-7B-Instruct, Training Type=RFT2025.0982.8
RISK-SFT-7BModel Category=Ours, Base Model=Qwen2.5-VL-7B-Instruct, Training Type=SFT2025.0975.3
GPT-4oModel Category=Commercial Models2025.0974
Qwen2.5-VL-72BModel Category=General Open-source Models2025.0967.8
Qwen-VL-MaxModel Category=Commercial Models2025.0950
Qwen2.5-VL-7BModel Category=General Open-source Models2025.090.6
InternVL3-8BModel Category=General Open-source Models2025.090
UI-TARS-2BModel Category=GUI-specific Models (SFT)2025.090
UI-TARS-7BModel Category=GUI-specific Models (SFT)2025.090
UI-TARS-72BModel Category=GUI-specific Models (SFT)2025.090
OS-Atlas-7BModel Category=GUI-specific Models (SFT)2025.090
ShowUI-2BModel Category=GUI-specific Models (SFT)2025.090
Aguvis-7BModel Category=GUI-specific Models (SFT)2025.090
GUI-R1-3BModel Category=GUI-specific Models (RL)2025.090
GUI-R1-7BModel Category=GUI-specific Models (RL)2025.090
InfiGUI-R1-3BModel Category=GUI-specific Models (RL)2025.090
GUI-G1-3BModel Category=GUI-specific Models (RL)2025.090
UI-TARS-1.5-7BModel Category=GUI-specific Models (RL)2025.090
UI-Venus-Navi-7BModel Category=GUI-specific Models (RL)2025.090