Spatial Relationship Reasoning on SPAR-Bench
61.9Accuracy (Avg)Ours-DTR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ours-DTRReasoning Mode=detect-then-reason inference2026.06 | 61.9 | 61.1 | 46.9 | 68.3 | |
| Ours-LORReasoning Mode=language-only reasoning2026.06 | 60.5 | 58.5 | 47.7 | 67.3 | |
| GAMSI_S1+S2SI Dataset=950K2026.05 | 59.9 | — | — | — | |
| GPT-5.2Model Type=Proprietary2026.07 | 52.6 | 43.7 | 41.5 | 66.4 | |
| GAMSI_S1SI Dataset=800K2026.05 | 49.6 | — | — | — | |
| SenseNova-SI-InternVL3-8BSI Dataset=800K2026.05 | 48.9 | — | — | — | |
| VST2026.06 | 48.9 | 53.3 | 25.4 | 53.7 | |
| VST-7B-SFTSI Dataset=4M2026.05 | 47.2 | — | — | — | |
| OmniView-Space-R1†Mode=Tool-Integrated2026.07 | 44.8 | 44.1 | 38.1 | 47.8 | |
| Gemini-3-ProModel Type=Proprietary2026.07 | 43.1 | 40.6 | 21.7 | 58 | |
| OmniView-Space-R1‡Mode=Tool-Distillation2026.07 | 42.3 | 43.2 | 34.3 | 44.2 | |
| SenseNova-SI-Qwen3-VL-8BSI Dataset=8M2026.05 | 41.1 | — | — | — | |
| Qwen3-VL-8B-Instruct2026.05 | 39.9 | — | — | — | |
| GPT-4.1Model Type=Proprietary2026.07 | 39.9 | 38.2 | 37 | 43.1 | |
| Qwen3-VL-8BModel Scale=8B2026.06 | 39.6 | 35.7 | 30 | 46.7 | |
| OmniView-Space-ZeroMode=Tool-Integrated2026.07 | 39.6 | 33.5 | 35.8 | 46.3 | |
| SpaceR2026.06 | 39.2 | 32.3 | 33 | 47.6 | |
| Bagel-7B-MoT2026.05 | 39.1 | — | — | — | |
| VST-3B-SFTSI Dataset=4M2026.05 | 38.1 | — | — | — | |
| Cambrian-S-7BSI Dataset=590K2026.05 | 38 | — | — | — | |
| Cambrian-S-7BModel Type=Spatial Intelligence2026.07 | 37.9 | 29.7 | 31.3 | 49.6 | |
| ViLaSR-7BModel Type=Spatial Intelligence2026.07 | 37.4 | 35.3 | 29 | 42.5 | |
| SpaceR-7BSI Dataset=151K2026.05 | 37.3 | — | — | — | |
| ViLaSR-7BSI Dataset=81K2026.05 | 37.3 | — | — | — | |
| InternVL3-8BModel Type=Open-source General2026.07 | 37.3 | 28.1 | 39.3 | 45 | |
| Qwen3-VL-8B-InstructModel Type=Open-source General2026.07 | 36.5 | 35.6 | 32.9 | 39.6 | |
| GPT-4oAccess=Proprietary2026.05 | 36.4 | 29.3 | 24.9 | 45.1 | |
| InternVL3-8B2026.05 | 35.8 | — | — | — | |
| Qwen3-VL-4B-InstructModel Type=Open-source General2026.07 | 35.6 | 30.8 | 33.1 | 40.5 | |
| LLaVA-OneVision-1.5-8BModel Scale=1.5-8B2026.06 | 35.5 | 31.9 | 26 | 42.4 | |
| Spatial-MLLM-4BSI Dataset=135K2026.05 | 35.3 | — | — | — | |
| SpatialLadder2026.06 | 34.4 | 25.8 | 33.2 | 42.6 | |
| SpaceR-7BModel Type=Spatial Intelligence2026.07 | 34.2 | 27 | 23.9 | 44.6 | |
| Qwen2.5-VL-7B-Instruct2026.05 | 33.8 | — | — | — | |
| Qwen2.5-VL-7B-InstructModel Type=Open-source General2026.07 | 33.8 | 27.6 | 27.6 | 41.8 | |
| SR-3D (Base)Model Variant=Base2026.06 | 33.4 | 24.1 | 37.6 | 40.1 | |
| Qwen3-VL-2B-Instruct2026.05 | 33.3 | — | — | — | |
| OmniView-Space-SFTMode=Tool-Distillation2026.07 | 33.1 | 26.6 | 32.5 | 38.9 | |
| SpatialLadder-3BSI Dataset=260K2026.05 | 32.9 | — | — | — | |
| Cambrian-S-3BSI Dataset=590K2026.05 | 32.9 | — | — | — | |
| SpatialLadder-3BModel Type=Spatial Intelligence2026.07 | 32.9 | 24.9 | 29.4 | 42.4 | |
| NVILA-8B-Lite (Base)Model Scale=8B, Model Variant=Lite (Base)2026.06 | 32.3 | 24.9 | 28.6 | 40.3 | |
| LLaVA-OneVision-7BParameters=7B2026.05 | 31.2 | 21.8 | 26.1 | 40.1 | |
| InternVL-2.5-4BParameters=4B2026.05 | 30.6 | 25.7 | 29.8 | 35.2 | |
| Spatial-MLLM-4BModel Type=Spatial Intelligence2026.07 | 30.4 | 22.3 | 32.4 | 35.2 | |
| Qwen2.5-VL-7BParameters=7B2026.05 | 30.2 | 17.5 | 29.5 | 41.8 | |
| Qwen2.5-VL-7BModel Scale=7B2026.06 | 30.2 | 17.5 | 29.5 | 41.8 | |
| Qwen2.5-VL-3B-Instruct2026.05 | 30 | — | — | — | |
| InternVL2.5-8BModel Scale=8B2026.06 | 29.7 | 21.7 | 31.1 | 36.3 | |
| Kimi-VL-A3BParameters=3B2026.05 | 29.4 | 26.4 | 18.8 | 36.2 | |
| SpatialRGPT2026.06 | 28 | 26 | 22 | 32.1 | |
| CaMo-3BParameters=3B2026.05 | 27.6 | 13 | 31.3 | 42.6 | |
| InternVL3-2B2026.05 | 27.2 | — | — | — | |
| Qwen2.5-VL-3BParameters=3B2026.05 | 24.6 | 15.3 | 26.4 | 32.2 | |
| Cosmos-Reason1-7BModel Scale=7B2026.06 | 22.8 | 23.4 | 19.4 | 24.2 | |
| ViGoRL2026.06 | 21.1 | 25 | 18.1 | 18.6 | |
| MindCube-3BModel Type=Spatial Intelligence2026.07 | 20.8 | 16.8 | 22.6 | 23.9 |