Spatial reasoning on satellite imagery on SQuID (Tier 3)
18.84AccuracyQVLM
Evaluation Results
| Method | Links | |
|---|---|---|
| QVLMCode Generator=GPT-5, Segmentation Model=ConvNeXt2026.01 | 18.84 | |
| QVLMCode Generator=GPT-5, Segmentation Model=DINOv3 + Mask2Former2026.01 | 12.2 | |
| GPT-52026.01 | 10.83 | |
| QVLMCode Generator=gpt-oss-120B, Segmentation Model=ConvNeXt2026.01 | 5.88 | |
| QVLMCode Generator=Llama-3.1-8B, Segmentation Model=ConvNeXt2026.01 | 5.79 | |
| QWEN 30B A3B thinking2026.01 | 3.71 |