Skyfall AI has opened a waitlist for SKY-001, a spatial world model for architects, mechanical designers, and manufacturing teams who still build complex products and buildings as long sequences of sketches, extrusions, and room layouts. From an image of an engineered product, it produces a hierarchy of assemblies and parts, then the sketches and operations that build each part. From a building outline and a short brief, it lays out units and rooms and draws the floor plan one state at a time. The same intent can generate several candidate designs before one is taken apart. On construction benchmarks Skyfall ran itself, the speed and cost gaps versus the language models it tested are large, and the quality gap is small. Across 1,000 CAD components sampled from objects such as a Jaguar E-Type Roadster, a FANUC ArcMate 100iB robotic arm, and a jet engine, SKY-001 took 3.9 seconds and $0.005 per case at 79.7% mean IoU (intersection over union), against 199 seconds, $0.75, and 79.1% for GPT-6 Astra. Across 1,000 floor plans, it took 1.22 seconds and $0.0024 at 43.6% IoU, against 144.6 seconds, $0.446, and 40.6% for the same model. That is about 51 times faster and 150 times cheaper on the CAD set, and about 119 times faster and 186 times cheaper on the BIM set. The release comes from Skyfall, headquartered in San Francisco with a research hub in Toronto, and co-founded by Sam Pasupalak, Kaheer Suleman, and Sumit Pasupalak, who previously built Maluuba, later acquired by Microsoft. Pasupalak says the model is 18 months of work from that deep learning group, and presents it as a way to plan a physical design before it is built. The published system produces a visual design plan. Turning those sketches into an editable CAD feature history, and turning floor plans into walls, doors, and rooms an architect can keep editing, are the next steps the company has named.