Video Spatial Reasoning on SpaceVista-Bench
81.3AccuracyHuman
Evaluation Results
| Method | Links | |
|---|---|---|
| Human2025.10 | 81.3 | |
| SpaceVista-7BParameters=7B, RL=true2025.10 | 39.2 | |
| SpaceVista-7BParameters=7B2025.10 | 35.4 | |
| Gemini-2.5-pro2025.10 | 33.8 | |
| GPT-52025.10 | 33.7 | |
| Qwen2.5-VL-72BParameters=72B2025.10 | 31.1 | |
| InternVL3.5-38BParameters=38B2025.10 | 30.7 | |
| Qwen2.5-VL-7BFine-tuned on=SpaceVista-1M2025.10 | 29.5 | |
| Qwen2.5-VL-7BParameters=7B2025.10 | 28.9 | |
| VG LLM-4BParameters=4B2025.10 | 28.8 | |
| InternVL3.5-8BParameters=8B2025.10 | 24.5 | |
| SpatialMLLM-4BParameters=4B2025.10 | 24.2 | |
| LLaVA-NeXT-Video-7BParameters=7B2025.10 | 23.7 | |
| VILASR-7BParameters=7B2025.10 | 23.6 | |
| SpaceR-7BParameters=7B2025.10 | 21.2 | |
| LLAVA-Onevision-7BParameters=7B2025.10 | 13.6 |