List your company

Web Navigation on Mind2Web

58.7Overall Success RateTriton-GRPO-32B
-2.34813.50129.3545.199Dec 14, 2023May 8, 2024Oct 1, 2024Feb 25, 2025Jul 21, 2025Dec 14, 2025May 10, 2026

Evaluation results

MethodLinks
Triton-GRPO-32BSize=32B2026.0458.7———70.1—
CogAgentRepresentations of screen inputs=Image, Element selection candidates=top-502023.1258.262.35459.4——
LLaMA2-70BRepresentations of screen inputs=HTML, Element selection candidates=top-502023.1254.455.851.655.7——
Triton-ORPO-32BSize=32B2026.0453.2———63.4—
LLaMA2-7BRepresentations of screen inputs=HTML, Element selection candidates=top-502023.1250.152.747.150.3——
Triton-SFT-32BSize=32B2026.0447.6———60.5—
Flan-T5XLRepresentations of screen inputs=HTML, Element selection candidates=top-102023.1243.55238.939.6——
GPT-4.5-preview-2025-02-272026.0442.4———59.8—
Gemini-2.5-pro-preview-05-062026.0441.5———59.1—
Claude-4-5-noThink2026.0441.4———58.8—
Grok-32026.0441.1———58.5—
o4-mini-2025-04-162026.0439.2———56.8—
Gemini-2.5-Flash-preview-04-172026.0438———55.3—
Grok-3-fast2026.0436.6———54.2—
Gemini-2.0-Flash2026.0435———53.1—
GPT-4.1-mini-2025-04-142026.0433.2———52—
GPT-4Representations of screen inputs=HTML, Few-shot=true, Element selection candidates=top-102023.1230.936.230.126.4——
GPT-4.1-nano-2025-04-142026.0428.4———48.4—
DeepSeek-V3Size=37/671B2026.0425.2———43.4—
CogVLMRepresentations of screen inputs=Image, Element selection candidates=top-502023.1223.937.123.426.3——
Mem-W-8BBackbone=UI-Venus-1.5-8B2026.0523.53—————
Mem-W-4BBackbone=Qwen3-VL-4B2026.0522.55—————
Mem-W-7BBackbone=UI-TARS-1.5-7B2026.0520.59—————
MAI-UI-2BModel Class=Open-source Native Models2026.0519.41—————
Qwen3-32B-Instruct ChatSize=32B2026.0417.7———36.9—
GPT-3.5Representations of screen inputs=HTML, Few-shot=true, Element selection candidates=top-102023.1217.418.617.416.2——
Qwen3-VL-8BModel Class=Open-source Native Models2026.0513.73—————
Qwen3-VL-32BModel Class=Open-source Native Models2026.0513.73—————
Qwen3-VL-4BModel Class=Mem-W Models2026.0513.72—————
Step-GUI-4BModel Class=Open-source Native Models2026.0512.75—————
Qwen3-4B-Instruct ChatSize=4B2026.0411.2———29.7—
GLM-4.1V-9B-ThinkingModel Class=Open-source Native Models2026.0510.78—————
Qwen-VLRepresentations of screen inputs=Image, Element selection candidates=top-502023.1210.212.610.18——
Qwen2.5-VL-7BModel Class=Open-source Native Models2026.059.8—————
Qwen3-VL-2BModel Class=Open-source Native Models2026.059.8—————
GUI-Owl-1.5-8BModel Class=Open-source Native Models2026.058.82—————
Qwen3-1.7B-Instruct ChatSize=1.7B2026.046.4———23.9—
UI-TARS-1.5-7BModel Class=Mem-W Models2026.055.88—————
UI-Venus-1.5-8BModel Class=Mem-W Models2026.055.88—————
UI-Venus-1.5-30B-A3BModel Class=Open-source Native Models2026.053.92—————
Qwen3-0.6B-Instruct ThinkSize=0.6B2026.040———3.5—
CogAgentbase_agent=CogAgent2026.06—————41.2
CogAgent + VisCriticbase_agent=CogAgent, critic_module=VisCritic2026.06—————44.7
Qwen2.5-VL Agentbase_agent=Qwen2.5-VL Agent2026.06—————47.3
Qwen2.5-VL Agent + GUI-Critic-R1base_agent=Qwen2.5-VL Agent, critic_module=GUI-Critic-R12026.06—————51.8
Qwen2.5-VL Agent + GUI-PRAbase_agent=Qwen2.5-VL Agent, critic_module=GUI-PRA2026.06—————50.1
Qwen2.5-VL Agent + GuidNavbase_agent=Qwen2.5-VL Agent, critic_module=GuidNav2026.06—————51.2
Qwen2.5-VL Agent + VisCriticbase_agent=Qwen2.5-VL Agent, critic_module=VisCritic2026.06—————54.1
SeeClickbase_agent=SeeClick2026.06—————33.4
SeeClick + BacktrackAgentbase_agent=SeeClick, critic_module=BacktrackAgent2026.06—————36.9
SeeClick + GUI-Critic-R1base_agent=SeeClick, critic_module=GUI-Critic-R12026.06—————37.2
SeeClick + GUI-PRAbase_agent=SeeClick, critic_module=GUI-PRA2026.06—————36.2
SeeClick + VisCriticbase_agent=SeeClick, critic_module=VisCritic2026.06—————37.6
ShowUIbase_agent=ShowUI2026.06—————43.5
ShowUI + GUI-PRAbase_agent=ShowUI, critic_module=GUI-PRA2026.06—————46.2
ShowUI + VisCriticbase_agent=ShowUI, critic_module=VisCritic2026.06—————48.9