N3D-VLM: Native 3D Grounding Enables Accurate Spatial Reasoning in Vision-Language Models Paper • 2512.16561 • Published Dec 18, 2025 • 20
