Spatial Reasoning on VSI-Bench tiny
51.61Avg ScoreThink3D
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Think3DBackbone=Gemini-2.5-Pro, Exploration Iterations=22026.01 | 51.61 | 46.93 | 37.3 | 54 | 68.24 | |
| Think3DBackbone=GPT-4.1, Exploration Iterations=22026.01 | 51.14 | 45.26 | 45.3 | 46 | 68 | |
| VLM-3RModel Type=Specialized Spatial Models2026.01 | 51.08 | 46.94 | 64.27 | 38 | 55.1 | |
| CaMo-3BTraining=SFT2026.05 | 51 | 49 | 40 | 46 | 69.4 | |
| GLM-4.5VModel Type=Proprietary models2026.01 | 48.72 | 34.69 | 41.03 | 40 | 79.16 | |
| SpaceR-3BTraining=GRPO2026.05 | 48.5 | 38.8 | 36 | 44 | 75.5 | |
| GPT-4.12026.01 | 48.18 | 40.8 | 40.63 | 43.3 | 68 | |
| Spatial-MLLMModel Type=Specialized Spatial Models2026.01 | 46.94 | 38.3 | 44 | 40 | 65.31 | |
| Think3DBackbone=Qwen3-VL-4BRL, Training=Think3D-RL, Exploration Iterations=32026.01 | 46.33 | 39.46 | 44 | 42.67 | 59.86 | |
| Gemini-2.5-Pro2026.01 | 45.16 | 45.58 | 28.67 | 50.67 | 55.73 | |
| Qwen2.5-VL-3BTraining=SFT2026.05 | 44.4 | 44.9 | 38 | 38 | 57.1 | |
| Doubao-1.5Model Type=Proprietary models2026.01 | 43.07 | 42.86 | 18 | 40 | 71.4 | |
| REVPTModel Type=Specialized Spatial Models2026.01 | 39.9 | 28.57 | 40 | 40 | 51.02 | |
| Qwen3-VL-4BRLTraining=Think3D-RL2026.01 | 39.37 | 26.53 | 44 | 40 | 46.94 | |
| Think3DBackbone=Qwen3-VL-4B, Exploration Iterations=32026.01 | 39.08 | 30.61 | 44 | 29.33 | 52.38 | |
| Internvl3-78BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 38.3 | — | — | — | — | |
| Qwen3-VL-4B2026.01 | 38.28 | 34.69 | 40.67 | 35.33 | 42.44 | |
| SpaceVista-7bModel Category=Open-Source Specialized Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 35.3 | — | — | — | — | |
| Qwen3-VL-4BGRPOTraining=GRPO2026.01 | 33.3 | 28.57 | 38 | 36 | 30.61 | |
| Gemini-2.5-proModel Category=Closed-sourced Commercial Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 33 | — | — | — | — | |
| SpatialLadder-3BTraining=GRPO2026.05 | 32.8 | 30.6 | 30 | 32 | 38.8 | |
| GPT-5Model Category=Closed-sourced Commercial Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 32.3 | — | — | — | — | |
| GLM-4.1V-ThinkingModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 30.7 | — | — | — | — | |
| Internvl3.5-38BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 29.3 | — | — | — | — | |
| Internvl3.5-14BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 27.7 | — | — | — | — | |
| Qwen2.5VL-72BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 27.7 | — | — | — | — | |
| Claude-Sonnet-4Model Category=Closed-sourced Commercial Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 27.3 | — | — | — | — | |
| Qwen2.5VL-32BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 25.3 | — | — | — | — | |
| LLAVA-Onevision-72BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 25 | — | — | — | — | |
| RoBoBrainModel Type=Specialized Spatial Models2026.01 | 23.2 | 28.57 | 36 | 16 | 12.24 | |
| GLM-4.5VModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 23 | — | — | — | — | |
| GPT-4oModel Category=Closed-sourced Commercial Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 21.7 | — | — | — | — | |
| Claude-Opus-4.1Model Category=Closed-sourced Commercial Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 21.7 | — | — | — | — | |
| Gemini-2.5-flashModel Category=Closed-sourced Commercial Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 20.7 | — | — | — | — | |
| Internvl3-38BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 18.7 | — | — | — | — | |
| LLAVA-Onevision-7BModel Category=Open-Source General Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 17.5 | — | — | — | — | |
| Spatial-MLLMModel Category=Open-Source Specialized Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 17.3 | — | — | — | — | |
| VLM-3RModel Category=Open-Source Specialized Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 15.1 | — | — | — | — | |
| SpaceRModel Category=Open-Source Specialized Models, Instruction-tuned=true, Resolution=Same, FPS=Same2025.10 | 12.9 | — | — | — | — |