Static Spatial Reasoning on STI-Bench Static
37.1Avg. ScoreGemini-2.5-Pro
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-2.5-ProType=Proprietary Model (API)2026.06 | 37.1 | 34.2 | 53.4 | 32.3 | |
| Claude-3.7-SonnetType=Proprietary Model (API)2026.06 | 37 | 31.8 | 49 | 36.3 | |
| Gemini-2.0-FlashType=Proprietary Model (API)2026.06 | 36.9 | 33.7 | 50 | 33.7 | |
| ReReBackbone=INTERNVL2.5-8B2026.06 | 34.8 | 33.2 | 49.3 | 29.7 | |
| ReReBackbone=QWEN3-VL-4B2026.06 | 34.4 | 33.6 | 48.6 | 28.7 | |
| INTERNVL2.5-8BFramework=Baseline2026.06 | 32.1 | 30.1 | 45.9 | 27.8 | |
| GPT-4oType=Proprietary Model (API)2026.06 | 31 | 24.9 | 49.6 | 28.1 | |
| ReReBackbone=QWEN3-VL-8B2026.06 | 30.9 | 29.1 | 44.5 | 26.2 | |
| ReReBackbone=INTERNVL2.5-4B2026.06 | 30.6 | 22.2 | 43.2 | 32.5 | |
| INTERNVL2.5-4BFramework=Baseline2026.06 | 30.5 | 25.9 | 43.1 | 29 | |
| ReReBackbone=QWEN3-VL-2B2026.06 | 30.2 | 24.6 | 50 | 26.2 | |
| QWEN3-VL-4BFramework=Baseline2026.06 | 29.7 | 29.8 | 41.8 | 24.3 | |
| QWEN3-VL-8BFramework=Baseline2026.06 | 27.9 | 29.4 | 43.8 | 19.2 | |
| ReReBackbone=INTERNVL3-8B2026.06 | 27.8 | 28 | 37 | 23.3 | |
| ReReBackbone=INTERNVL3-2B2026.06 | 26.7 | 22.5 | 43.8 | 22.7 | |
| INTERNVL3-8BFramework=Baseline2026.06 | 24.5 | 20.1 | 44.5 | 19.2 | |
| INTERNVL3-2BFramework=Baseline2026.06 | 22.6 | 20.1 | 28.8 | 22.1 | |
| QWEN3-VL-2BFramework=Baseline2026.06 | 22.2 | 18 | 31.5 | 21.8 |