List your company

GUI Agent Task Success on AndroidWorld (online)

48.7Task Success RateGPT-5
1.69213.89626.138.304Feb 25, 2026

Evaluation results

MethodLinks
GPT-5Model Type=Agent Frameworks, Additional Module=UGround-v1-7B, Steps=202026.0248.7
GUI-Libra-4BModel Type=Native Models (single VLM), Steps=202026.0242.6
GUI-Libra-8BModel Type=Native Models (single VLM), Steps=202026.0242.6
GPT-4oModel Type=Agent Frameworks, Additional Module=UGround-v1-7B, Steps=202026.0242.6
GPT-5-miniModel Type=Agent Frameworks, Additional Module=UGround-v1-7B, Steps=202026.0240.9
Qwen3-VL-8BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.0239.1
GPT-4.1Model Type=Agent Frameworks, Additional Module=UGround-v1-7B, Steps=202026.0237.4
Qwen3-VL-4BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.0236.5
Qwen3-VL-32BModel Type=Native Models (single VLM), Steps=202026.0234.8
Qwen2.5-VL-72BModel Type=Native Models (single VLM), Steps=202026.0232.2
ScaleCUA-32BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.0230.6
Qwen3-VL-8B (Baseline)Model Type=Native Models (single VLM), Steps=202026.0230.4
Qwen2.5-VL-32BModel Type=Native Models (single VLM), Steps=202026.0229.6
GUI-Libra-7BModel Type=Native Models (single VLM), Steps=202026.0229.6
ScaleCUA-7BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.0227.2
Qwen3-VL-4B (Baseline)Model Type=Native Models (single VLM), Steps=202026.0227
GUI-Libra-3BModel Type=Native Models (single VLM), Steps=202026.0225.2
ScaleCUA-3BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.0223.7
GLM-4.1V-9B-ThinkingModel Type=Agent Frameworks, Additional Module=UGround-v1-7B, Steps=202026.0220.9
GLM-4.1V-9B-ThinkingModel Type=Native Models (single VLM), Steps=202026.0218.3
UI-TARS-1.5-7BModel Type=Native Models (single VLM), Steps=202026.0216.5
Qwen2.5-VL-7BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.0215.7
Qwen2.5-VL-7B (Baseline)Model Type=Native Models (single VLM), Steps=202026.027.8
Qwen2.5-VL-3BModel Type=Agent Frameworks, Additional Module=Step-wise Summary, Steps=202026.027
Qwen2.5-VL-3B (Baseline)Model Type=Native Models (single VLM), Steps=202026.023.5