Webpage Question Answering on VisualWebBench MultiUI-WQA
89.47AccuracyGPT-5-nano
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-5-nanoModel Category=Proprietary Models2025.10 | 89.47 | |
| COGSApproach Type=Data Synthesis Approaches2025.10 | 88.04 | |
| MultiUI-WQAApproach Type=Data Synthesis Approaches2025.10 | 86.6 | |
| Decompositional CoTApproach Type=Inference-time decomposition2025.10 | 86.12 | |
| Qwen2.5-VL-7B (base model)Model Category=Opensource Models (~7B)2025.10 | 85.65 | |
| Gemini 2.5 Flash-LiteModel Category=Proprietary Models2025.10 | 81.85 | |
| GPT-4o-miniModel Category=Proprietary Models2025.10 | 81.34 | |
| Claude Haiku 3.5Model Category=Proprietary Models2025.10 | 80.86 | |
| InternVL3.5-GPT-OSSModel Category=Opensource Models (~7B)2025.10 | 74.64 | |
| Phi-4-14BModel Category=Opensource Models (~7B)2025.10 | 74.16 | |
| UiX-Qwen2Model Category=Specialist Models2025.10 | 68.9 |