Spatial Reasoning on SQuID Tier 2
54.06AccuracyQVLM
Evaluation Results
| Method | Links | |
|---|---|---|
| QVLMCode Generator=GPT-5, Segmentation Model=ConvNeXt2026.01 | 54.06 | |
| QVLMCode Generator=gpt-oss-120B, Segmentation Model=ConvNeXt2026.01 | 47.62 | |
| QVLMCode Generator=Llama-3.1-8B, Segmentation Model=ConvNeXt2026.01 | 41.88 | |
| QVLMCode Generator=GPT-5, Segmentation Model=DINOv3 + Mask2Former2026.01 | 40.22 | |
| QWEN 30B A3B thinking2026.01 | 36.85 | |
| GPT-52026.01 | 34.09 |