List your company
ResearchFormerly WizWand

3D Spatial Reasoning on 3DSRBench

95.7AccuracyHuman Estimate
17.90838.10458.378.496Apr 29, 2025Jul 3, 2025Sep 6, 2025Nov 10, 2025Jan 14, 2026Mar 20, 2026May 25, 2026
Updated 3mo ago

Evaluation results

MethodLinks
Human Estimate2025.0495.7
GPT-5-0807Model Type=Proprietary2025.1168.2
Emb. TextProtocol=CoT2026.0167
Rotation TokensProtocol=Direct2026.0167
Qwen3.5Parameters=35BA3B, Internal Reasoning (Think mode)=false2026.0566.6
3DThinker2025.1065.6
SenseNova-U1Parameters=8B, Internal Reasoning (Think mode)=true2026.0564.88
Emb. Tokens (ViTPose)Protocol=CoT2026.0164
Rotation TextProtocol=CoT2026.0164
SenseNova-U1Parameters=30BA3B, Internal Reasoning (Think mode)=true2026.0562.96
w/ Entropy-RegularizationEvaluation Protocol=RL Fine-tuning2026.0562.9
SpatialThinker-30B (Ours)Model Scale=30B, Training Dataset=STVQA-7K2025.1162.1
Claude-4-Sonnet-0514Model Type=Proprietary2025.1161.9
Qwen3-VL-30BModel Scale=30B, Model Type=Open-Source General2025.1160.4
SpatialReasoner2025.1060.3
SpatialReasonerType=Specialist Reference2026.0560.3
Rotation TextProtocol=Direct2026.0160
w/ Tool-Encourage RewardEvaluation Protocol=RL Fine-tuning2026.0559.9
vanilla RFT (Base)Evaluation Protocol=RL Fine-tuning2026.0559.2
Emb. TextProtocol=Direct2026.0158
Emb. Tokens (COCO)Protocol=CoT2026.0158
SpaRE-7BModel size=7B2025.0457.5
Qwen3.5Parameters=9B, Internal Reasoning (Think mode)=false2026.0556.77
SpatialThinker2025.1056.4
SpatialThinker-7B (Ours)Model Scale=7B, Training Dataset=STVQA-7K2025.1156.4
Rotation TokensProtocol=CoT2026.0156
SpatialReasoner-R12025.1055.7
Qwen3VLParameters=30BA3B, Internal Reasoning (Think mode)=true2026.0555.55
Qwen2.5-VL-7B + Vanilla GRPOModel Scale=7B, Training Protocol=Vanilla GRPO, Training Dataset=STVQA-7K2025.1154.7
Mini-o3Mode=Zero-shot, Type=Visual Agents2026.0554.5
Qwen3VLParameters=8B, Internal Reasoning (Think mode)=true2026.0554.48
SpaRE-2BModel size=2B2025.0454.4
Gemma4Parameters=26BA4B, Internal Reasoning (Think mode)=false2026.0553.61
Qwen2.5-VL-7B + SFTModel Scale=7B, Training Protocol=SFT, Training Dataset=STVQA-7K2025.1153.6
InternVL2-8BModel size=8B2025.0453.3
SpatialThinker-3B (Ours)Model Scale=3B, Training Dataset=STVQA-7K2025.1152.9
VLAA-Thinker-Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Model Type=Open-Source General2025.1152.2
SpaceOmModel Type=Open-Source Spatial2025.1152.2
DeepEyesMode=Zero-shot, Type=Visual Agents2026.0551.6
LLaVA-NeXT-8BModel size=8B2025.0451.1
SpaceThinkerModel Type=Open-Source Spatial2025.1151.1
Qwen2.5-VL-3B + SFTModel Scale=3B, Training Protocol=SFT, Training Dataset=STVQA-7K2025.1150.8
Qwen2.5-VL-3B + Vanilla GRPOModel Scale=3B, Training Protocol=Vanilla GRPO, Training Dataset=STVQA-7K2025.1150.1
Qwen2VL-7BModel size=7B2025.0449.2
Emb. Tokens (COCO)Protocol=Direct2026.0149
Emb. Tokens (ViTPose)Protocol=Direct2026.0149
Qwen2VL-2BModel size=2B2025.0448.8
Qwen2.5-VL 7BType=Generalist Baseline2026.0548.4
Qwen2.5-VL-7BModel Scale=7B, Model Type=Open-Source General2025.1148.4
LLaVA-NeXT-8BModel Scale=8B, Model Type=Open-Source General2025.1148.4
Spatial-RGPT-7B w/ depthModel Scale=7B, Model Type=Open-Source Spatial, Auxiliary Inputs=depth2025.1148.4
Claude-3.5-Sonnet-0620Model Type=Proprietary2025.1148.2
SATORI-R1Model Type=Open-Source Spatial2025.1148
SpaceLLaVa2025.0447.2
InternVL2-2BModel size=2B2025.0446.7
GPT-4o2025.0445.3
GPT-4o-0513Model Type=Proprietary2025.1144.3
Qwen2.5-VL-3BModel Scale=3B, Model Type=Open-Source General2025.1144
Cambrian-1-8BModel Scale=8B, Model Type=Open-Source General2025.1142.2
SpaceLLaVA-13BModel Scale=13B, Model Type=Open-Source Spatial2025.1142
SpatialBot-3BModel Scale=3B, Model Type=Open-Source Spatial2025.1141.1
LLaVAModel Scale=1.5-13B2026.0140
GPT-4o-mini2025.0439.1
Random2025.0420.9