Spatial Reasoning on ReasonMap Plus
89.7AccuracyGPT-5
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5Model Type=Commercial2026.06 | 89.7 | |
| Gemini-2.5-FlashModel Type=Commercial2026.06 | 85.4 | |
| Gemini-2.5-ProModel Type=Commercial2026.06 | 81.3 | |
| Qwen3-235BModel Type=Open-Source2026.06 | 76.7 | |
| PERIA-8BModel Type=Proposed2026.06 | 72.6 | |
| w/o ToolModel Type=Ablation, Study=No Tools2026.06 | 72.4 | |
| Qwen3-32BModel Type=Open-Source2026.06 | 68.8 | |
| Qwen3-30BModel Type=Open-Source2026.06 | 67.8 | |
| w/o RLModel Type=Ablation, Study=SFT only2026.06 | 66.2 | |
| Qwen3-8BModel Type=Backbone Baseline2026.06 | 65.7 | |
| w/o SFTModel Type=Ablation, Study=RL only2026.06 | 64.2 | |
| InternVL3.5-8BModel Type=Open-Source2026.06 | 53.3 | |
| R1-OneVision-7BModel Type=Tool-Augmented2026.06 | 52.1 | |
| InternVL3.5-14BModel Type=Open-Source2026.06 | 51.3 | |
| Mini-o3Model Type=Tool-Augmented2026.06 | 45.9 | |
| VTool-R1-7BModel Type=Tool-Augmented2026.06 | 38.3 |