List your company

GUI Agent Interaction on OSWorld (Step Success Rate)

42.9Success Rate (Max Steps: 15)Claude 4.5 Sonnet
1.40412.17722.9533.723May 28, 2026

Evaluation results

MethodLinks
Claude 4.5 SonnetModel Category=Proprietary Models, Data Type=In-house2026.0542.9—58.1—
RoTS-32BModel Category=Open-Weights Models (Larger Size), Data Type=Open-source2026.0542.833.847.4—
Qwen3-VL-PlusModel Category=Proprietary Models, Data Type=In-house2026.0533.124.535.2—
Qwen3-VL-FlashModel Category=Proprietary Models, Data Type=In-house2026.0532.122.141.6—
Doubao-1.5-ThinkingModel Category=Proprietary Models, Data Type=In-house2026.0531.9—40—
RoTS-7BModel Category=Open-Weights Models (Smaller Size), Data Type=Open-source2026.0531.726.336.3—
OpenCUA-32BModel Category=Open-Weights Models (Larger Size), Data Type=Open-source2026.0529.715.534.1—
Qwen3-VL-8B-ThinkingModel Category=Open-Weights Models (Smaller Size), Data Type=In-house2026.0529.221.633.9—
Qwen3-VL-32B-ThinkingModel Category=Open-Weights Models (Larger Size), Data Type=In-house2026.0528.121.141—
GUI-OWL-7BModel Category=Open-Weights Models (Smaller Size), Data Type=In-house2026.0527.114.729.4—
OpenAI CUAModel Category=Proprietary Models, Data Type=In-house2026.0526—31.3—
UI-TARS-1.5-7BModel Category=Open-Weights Models (Smaller Size), Data Type=In-house2026.0524.59.527.3—
OpenCUA-7BModel Category=Open-Weights Models (Smaller Size), Data Type=Open-source2026.0524.312.528.2—
UI-TARS-72B-DPOModel Category=Open-Weights Models (Larger Size), Data Type=In-house2026.05241125.8—
Qwen2.5-VL-72BModel Category=Open-Weights Models (Larger Size), Data Type=In-house2026.054.41.15—
Qwen2.5-VL-32BModel Category=Open-Weights Models (Larger Size), Data Type=In-house2026.0530.73.9—
CogAgentbase_agent=CogAgent2026.06———8.9
CogAgent + VisCriticbase_agent=CogAgent, critic_module=VisCritic2026.06———10.6
Qwen2.5-VL Agentbase_agent=Qwen2.5-VL Agent2026.06———14.2
Qwen2.5-VL Agent + GUI-Critic-R1base_agent=Qwen2.5-VL Agent, critic_module=GUI-Critic-R12026.06———17.4
Qwen2.5-VL Agent + GUI-PRAbase_agent=Qwen2.5-VL Agent, critic_module=GUI-PRA2026.06———16.3
Qwen2.5-VL Agent + GuidNavbase_agent=Qwen2.5-VL Agent, critic_module=GuidNav2026.06———17
Qwen2.5-VL Agent + VisCriticbase_agent=Qwen2.5-VL Agent, critic_module=VisCritic2026.06———17.1
SeeClickbase_agent=SeeClick2026.06———6.8
SeeClick + BacktrackAgentbase_agent=SeeClick, critic_module=BacktrackAgent2026.06———9
SeeClick + GUI-Critic-R1base_agent=SeeClick, critic_module=GUI-Critic-R12026.06———9.3
SeeClick + GUI-PRAbase_agent=SeeClick, critic_module=GUI-PRA2026.06———8.2
SeeClick + VisCriticbase_agent=SeeClick, critic_module=VisCritic2026.06———9.5
ShowUIbase_agent=ShowUI2026.06———10.3
ShowUI + GUI-PRAbase_agent=ShowUI, critic_module=GUI-PRA2026.06———12
ShowUI + VisCriticbase_agent=ShowUI, critic_module=VisCritic2026.06———12.8