Spatial Reasoning on MMSI-Bench (test)
52.8PR ScoreGCA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GCAParadigm=Tool-Integrated Spatial Agents2025.11 | 52.8 | 45 | 44.7 | — | 47.6 | |
| GPT-5.2Category=Proprietary2026.02 | 42.9 | 47.6 | 36 | 40.4 | 42 | |
| Gemini-3-ProCategory=Proprietary2026.02 | 42.9 | 50 | 38 | 40.9 | 42.7 | |
| Gemini-2.5-ProParadigm=Baseline Foundation VLMs2025.11 | 39 | 36.2 | 33.3 | — | 36.9 | |
| VILASRParadigm=Training-based Spatial VLMs2025.11 | 35.9 | 26 | 21 | — | 29.8 | |
| GLM-4.5VParadigm=Baseline Foundation VLMs2025.11 | 35.6 | 36.9 | 29.3 | — | 33.8 | |
| Qwen3-VL-ThinkingParadigm=Baseline Foundation VLMs2025.11 | 33.7 | 40 | 23.3 | — | 32.6 | |
| Qwen2.5-VL-72BCategory=Open-Weight2026.02 | 33.3 | 38.5 | 24.7 | 28.8 | 31.8 | |
| InternVL-2.5-8BCategory=Open-Weight2026.02 | 31 | 30 | 20.7 | 23.2 | 27.8 | |
| LLaVA-OneVision-4BCategory=Open-Weight2026.02 | 30.8 | 26.9 | 26 | 26.8 | 28.8 | |
| Video-R1Paradigm=Training-based Spatial VLMs2025.11 | 30.5 | 25.4 | 22 | — | 27.8 | |
| Video-R1Base Model=Qwen2.5-VL-7B2026.02 | 30.5 | 33.8 | 18.7 | 24.7 | 28 | |
| HATCHBase Model=Qwen2.5-VL-3B2026.02 | 30.5 | 20.8 | 23.3 | 24.7 | 27 | |
| SpatialLadderParadigm=Training-based Spatial VLMs2025.11 | 30.3 | 23.3 | 16 | — | 25.4 | |
| SpaceR-7BBase Model=Qwen2.5-VL-7B2026.02 | 30.3 | 25.4 | 22 | 21.7 | 26.7 | |
| VLaserParadigm=Training-based Spatial VLMs2025.11 | 29.8 | 26.9 | 26 | — | 27.3 | |
| GPT-4.1Category=Proprietary2026.02 | 29.3 | 33.8 | 38.7 | 31.3 | 31.7 | |
| SpaceRParadigm=Training-based Spatial VLMs2025.11 | 29.1 | 29.4 | 21.9 | — | 26.9 | |
| TIGeRParadigm=Tool-Integrated Spatial Agents2025.11 | 29.1 | 27.7 | 26 | — | 27.8 | |
| RoboBrain-2.0Paradigm=Training-based Spatial VLMs2025.11 | 28.9 | 28.8 | 22.5 | — | 28.9 | |
| InternVL-2.5-4BCategory=Open-Weight2026.02 | 28.7 | 26.2 | 20.7 | 28.3 | 27.1 | |
| Spatial-MLLMParadigm=Training-based Spatial VLMs2025.11 | 28.5 | 25.4 | 18 | — | 26.1 | |
| Qwen2.5-VL-7BBase Model=Qwen2.5-VL-7B2026.02 | 28.4 | 21.5 | 29.3 | 25.8 | 27.1 | |
| GPT-4oParadigm=Baseline Foundation VLMs2025.11 | 28 | 32.3 | 36 | — | 30.3 | |
| Spatial-MLLM-4BBase Model=Qwen2.5-VL-3B2026.02 | 28 | 21.5 | 14.7 | 22.7 | 24.1 | |
| Qwen2.5-VL-3BBase Model=Qwen2.5-VL-3B2026.02 | 27.8 | 19.2 | 23.3 | 25.8 | 25.6 | |
| Qwen2.5-VL-32BCategory=Open-Weight2026.02 | 25.9 | 26.9 | 28.7 | 27.3 | 26.7 | |
| SpatialLLMParadigm=Training-based Spatial VLMs2025.11 | 24.5 | 23.1 | 22.7 | — | 25.3 | |
| SpatialLadder-3BBase Model=Qwen2.5-VL-3B2026.02 | 24.1 | 23.1 | 26.7 | 20.2 | 23.6 |