List your company

Spatial Reasoning on BLINK Multi-view (test)

63.91AccuracyThink3D (GPT-4.1)
35.736443.050750.36557.6793Jan 19, 2026

Evaluation results

MethodLinks
Think3D (GPT-4.1)Framework=Think3D, Backbone=GPT-4.12026.0163.91
Spatial-MLLM2026.0156.06
RoBoBrain2026.0155.64
Think3D (Qwen3-VL-4BRL)Framework=Think3D, Optimization Strategy=RL (Think3D-RL)2026.0153.64
Think3D (Gemini-2.5-Pro)Framework=Think3D, Backbone=Gemini-2.5-Pro2026.0152.88
Qwen3-VL-4BGRPOOptimization Strategy=GRPO2026.0152.38
REVPT2026.0151.89
Doubao-1.52026.0150.93
Think3D (Qwen3-VL-4B)Framework=Think3D, Backbone=Qwen3-VL-4B2026.0148.62
Qwen3-VL-4B2026.0147.87
Qwen3-VL-4BRLOptimization Strategy=RL (Think3D-RL)2026.0145.11
Gemini-2.5-Pro2026.0144.86
VLM-3R2026.0141.35
GLM-4.5V2026.0139.85
GPT-4.12026.0136.82