Video Spatial Reasoning on Video-RoboSpatial
92.1Configuration ScoreHuman Level
Evaluation results
| Method | Links | |
|---|---|---|
| Human LevelSize=–2026.03 | 92.1 | |
| Thinking with Spatial CodeSize=4B2026.03 | 67 | |
| Thinking with Spatial Code + 2D boxSize=4B2026.03 | 67 | |
| Thinking with Spatial Code (w/o RL)Size=4B2026.03 | 65.3 | |
| Thinking with Spatial Code (w/o RL) + 2D boxSize=4B2026.03 | 65.3 | |
| GPT-5Size=–2026.03 | 60.3 | |
| LLaVA-VideoSize=72B2026.03 | 58 | |
| SpaceRSize=7B2026.03 | 56 | |
| LLaVA-OneVisionSize=72B2026.03 | 56 | |
| Seed-1.6Size=230B2026.03 | 54.3 | |
| LLaVA-OneVisionSize=7B2026.03 | 54.3 | |
| Gemini-2.5-ProSize=–2026.03 | 53.3 | |
| GPT-4oSize=–2026.03 | 53 | |
| Qwen3-VLSize=4B2026.03 | 52.7 | |
| Qwen3-VL + 2D boxSize=4B2026.03 | 52.7 | |
| LLaVA-VideoSize=7B2026.03 | 52 | |
| Qwen2.5-VLSize=7B2026.03 | 49.7 | |
| SpatialLadderSize=3B2026.03 | 49.3 | |
| Spatial-MLLMSize=4B2026.03 | 49 | |
| Qwen3-VLSize=8B2026.03 | 48 |