Temporal Spatial Reasoning on VSIBench
62.9Average AccuracyG2F-RAG
Evaluation results
| Method | Links | |
|---|---|---|
| G2F-RAGSize=14B, Base Model=InternVL3.52026.04 | 62.9 | |
| InternVL3.5Size=14B2026.04 | 60.8 | |
| G2F-RAGSize=8B, Base Model=InternVL3.52026.04 | 60.1 | |
| Qwen3-VLSize=8B2026.04 | 59.4 | |
| G2F-RAGSize=4B, Base Model=InternVL3.52026.04 | 59 | |
| Qwen3-VLSize=4B2026.04 | 58.4 | |
| InternVL3.5Size=8B2026.04 | 56.3 | |
| InternVL3.5Size=4B2026.04 | 54.9 | |
| InternVL3Size=14B2026.04 | 48.9 | |
| Gemini1.5 ProSize=-2026.04 | 48.8 | |
| G2F-RAGSize=7B, Base Model=Qwen2.5-VL2026.04 | 44.8 | |
| G2F-RAGSize=7B, Base Model=LLaVA-Video2026.04 | 42.5 | |
| InternVL2.5Size=8B2026.04 | 41.6 | |
| InternVL3Size=8B2026.04 | 41.6 | |
| Qwen2.5-VLSize=7B2026.04 | 37.4 | |
| LLaVA-VideoSize=7B2026.04 | 35.7 | |
| GPT-4o (2024-05-13)Size=-2026.04 | 34 | |
| VILA-1.5Size=8B2026.04 | 28.9 | |
| Qwen2.5-VLSize=3B2026.04 | 27.9 |