A user demonstrated running GLM 5.3 and GLM 5.3 Flash locally on an RTX PRO 6000 WS workstation to build a penthouse scene in Blender via BlenderMCP. The Q4-quantized models required significant VRAM, with Flash needing approximately 190-200GB and the full GLM 5.3 requiring 450-470GB plus context headroom. The experiment showcases the feasibility of using large open-weight models for AI-driven 3D scene generation through Blender's Model Context Protocol integration.