Spatial Reasoning on VSI-Bench tiny (Detailed Metrics)
64Pass@1 Accuracy (Rel Dist)GPT-5.2
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.2AlloSpatial=Integrated, Setting=training-free, Input frames=322026.06 | 64 | 54 | — | 47.4 | 33.4 | 63.3 | 52.4 | — | 41.1 | 51 | — | — | 79.6 | |
| Claude-4.6-OpusAlloSpatial=Integrated, Setting=training-free, Input frames=322026.06 | 64 | 56 | — | 59 | 34.3 | 67.3 | 54.8 | — | 62.7 | 65.3 | — | — | 40.8 | |
| Gemini-3-ProAlloSpatial=Integrated, Setting=training-free, Input frames=322026.06 | 62 | 61 | — | 51.8 | 36.8 | 57.7 | 62.6 | — | 67.7 | 65.3 | — | — | 83.7 | |
| Gemini-3-ProModel Category=Proprietary Foundation Models, Input Frames=72026.06 | 56 | 45.2 | 32.2 | 29.4 | 12.6 | 38.3 | 48.4 | 58.3 | 52.7 | 49 | — | 75.5 | — | |
| Cambrian-S-3BModel Category=Specialized Spatial Models, Input Frames=72026.06 | 54 | 49.5 | 51.7 | 67.1 | 22.6 | 73.8 | 43.2 | 47.3 | 37.1 | 26.5 | — | 71.4 | — | |
| Gemini-3-ProAlloSpatial=None, Setting=training-free, Input frames=322026.06 | 54 | 55.2 | — | 47.8 | 32.1 | 71.3 | 55 | — | 44.8 | 57.1 | — | — | 79.6 | |
| Gemini-2.5-ProModel Category=Proprietary Foundation Models, Input Frames=72026.06 | 52 | 46.2 | 39.4 | 34.7 | 14.3 | 56.3 | 52.4 | 53 | 45.7 | 38.8 | — | 75.5 | — | |
| AlloSpatial-8BModel Category=Specialized Spatial Models, Input Frames=72026.06 | 52 | 54.2 | 47.8 | 46.1 | 26 | 65.6 | 53.4 | 60.6 | 64.1 | 46.9 | — | 79.6 | — | |
| AlloSpatial-4BModel Category=Specialized Spatial Models, Input Frames=72026.06 | 50 | 53.5 | 56.3 | 54.1 | 42.3 | 68.8 | 60 | 50.8 | 59.3 | 34.7 | — | 59.2 | — | |
| Qwen3-VL-32BModel Category=Open-source Foundation Models, Input Frames=72026.06 | 46 | 53.1 | 55.3 | 60.6 | 35.1 | 71.5 | 54 | 50.9 | 51.4 | 36.7 | — | 69.4 | — | |
| Think3D-4BModel Category=Specialized Spatial Models, Input Frames=72026.06 | 44.7 | — | — | — | — | — | — | 45.4 | 39 | 36.7 | — | 61.2 | — | |
| Spatial-MLLM-4BModel Category=Specialized Spatial Models, Input Frames=72026.06 | 44 | 49.1 | 53 | 76.5 | 27 | 57.9 | 50.6 | 45.2 | 40.9 | 38.8 | — | 57.1 | — | |
| GPT-5.2Model Category=Proprietary Foundation Models, Input Frames=72026.06 | 42 | 37.7 | 36.4 | 28.6 | 24.3 | 52.3 | 40.6 | 39 | 28.4 | 42.9 | 42.9 | — | — | |
| GPT-5.2AlloSpatial=None, Setting=training-free, Input frames=322026.06 | 42 | 46.7 | — | 52.5 | 34.9 | 67.5 | 50.6 | — | 40.7 | 34.7 | — | — | 51 | |
| Claude-4.6-OpusAlloSpatial=None, Setting=training-free, Input frames=322026.06 | 40 | 38.4 | — | 46.9 | 18.5 | 62.1 | 26.8 | — | 47.2 | 34.7 | — | — | 30.6 | |
| Gemma-4-E4BModel Category=Open-source Foundation Models, Input Frames=72026.06 | 38 | 33.5 | 30.1 | 19.6 | 24.5 | 33.1 | 43.2 | 36.9 | 40 | 30.6 | — | 38.8 | — | |
| InternVL3.5-4BModel Category=Open-source Foundation Models, Input Frames=72026.06 | 38 | 48.8 | 50.9 | 77.3 | 19.4 | 60.4 | 46.4 | 46.8 | 47.1 | 40.8 | — | 61.2 | — | |
| Qwen3-VL-4BModel Category=Open-source Foundation Models, Input Frames=72026.06 | 36 | 45.1 | 47.7 | 44.3 | 34.5 | 64.4 | 47.8 | 42.4 | 37.8 | 24.5 | — | 71.4 | — | |
| Qwen3-VL-8BModel Category=Open-source Foundation Models, Input Frames=72026.06 | 36 | 48.5 | 51.5 | 50.4 | 40.2 | 67.5 | 48 | 45.5 | 44.1 | 30.6 | — | 71.4 | — |