ScreenSpot-v2 (Web)
Multimodal · 2024-10-30
The web portion of ScreenSpot-v2: instructions on website screenshots with text and icon targets. A prediction is correct when the click point falls inside the target element's bounding box.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemma 4 12B IT | 92.9 |
| LFM3-VL-2B | 89.9 |
| Qwen3.5-4B | 89.7 |
| GPT-5.6 Luna | 89.0 |
| LFM2.5-VL-3B | 82.2 |
| Gemini 3.5 Flash-Lite | 78.3 |
| Qwen3.5-2B | 72.5 |
| Gemma 4 E2B IT | 52.9 |