Visual Probing on VisualProbe Easy
70.2AccuracyMini-o3
Evaluation Results
| Method | Links | |
|---|---|---|
| Mini-o3Model Type=Tool-Augmented2026.06 | 70.2 | |
| GPT-5Model Type=Commercial2026.06 | 62.4 | |
| Gemini-2.5-ProModel Type=Commercial2026.06 | 62.1 | |
| PERIA-8BModel Type=Proposed2026.06 | 61.7 | |
| Qwen3-235BModel Type=Open-Source2026.06 | 58.8 | |
| InternVL3.5-14BModel Type=Open-Source2026.06 | 58.1 | |
| Qwen3-32BModel Type=Open-Source2026.06 | 57.4 | |
| InternVL3.5-8BModel Type=Open-Source2026.06 | 54.7 | |
| Qwen3-8BModel Type=Backbone Baseline2026.06 | 54.6 | |
| Qwen3-30BModel Type=Open-Source2026.06 | 51.7 | |
| w/o RLModel Type=Ablation2026.06 | 51 | |
| Gemini-2.5-FlashModel Type=Commercial2026.06 | 46.1 | |
| R1-OneVision-7BModel Type=Tool-Augmented2026.06 | 39.2 | |
| w/o ToolModel Type=Ablation2026.06 | 36.1 | |
| VTool-R1-7BModel Type=Tool-Augmented2026.06 | 30.5 | |
| w/o SFTModel Type=Ablation2026.06 | 19.1 |