ScreenSpot-v2 (Desktop)
Multimodal · 2024-10-30
The desktop portion of ScreenSpot-v2: 334 instructions (194 text and 140 icon targets) on macOS and Windows screenshots. A prediction is correct when the click point falls inside the target element's bounding box.
Top models (higher is better)
| Model | Score |
|---|---|
| LFM3-VL-2B | 95.5 |
| GPT-5.6 Luna | 94.6 |
| Gemma 4 12B IT | 93.1 |
| Qwen3.5-4B | 91.6 |
| Gemini 3.5 Flash-Lite | 86.5 |
| LFM2.5-VL-3B | 78.7 |
| Qwen3.5-2B | 75.7 |
| Gemma 4 E2B IT | 52.1 |