G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning Paper • 2511.21688 • Published Nov 26, 2025 • 8 • 2