Spatial Reasoning on BabyVision
27.3AccuracyGemini-2.5-Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-2.5-ProModel Type=Commercial2026.06 | 27.3 | |
| GPT-5Model Type=Commercial2026.06 | 22.9 | |
| Qwen3-32BModel Type=Open-Source2026.06 | 22.9 | |
| Qwen3-235BModel Type=Open-Source2026.06 | 21.9 | |
| Gemini-2.5-FlashModel Type=Commercial2026.06 | 19.5 | |
| PERIA-8BModel Type=Proposed2026.06 | 15.5 | |
| Qwen3-30BModel Type=Open-Source2026.06 | 14.9 | |
| InternVL3.5-14BModel Type=Open-Source2026.06 | 14.6 | |
| R1-OneVision-7BModel Type=Tool-Augmented2026.06 | 14.4 | |
| InternVL3.5-8BModel Type=Open-Source2026.06 | 13.9 | |
| w/o ToolModel Type=Ablation2026.06 | 13.9 | |
| Qwen3-8BModel Type=Backbone Baseline2026.06 | 13 | |
| w/o RLModel Type=Ablation2026.06 | 12.4 | |
| VTool-R1-7BModel Type=Tool-Augmented2026.06 | 12.3 | |
| Mini-o3Model Type=Tool-Augmented2026.06 | 11.6 | |
| w/o SFTModel Type=Ablation2026.06 | 7.2 |