Skyfall AI describes SKY-001 for CAD and BIM design

Skyfall AI describes SKY-001 as the first world model for designing complex physical products, with an initial focus on design and engineering.[1][2] Skyfall says the model breaks high-level designs into smaller individual parts.[2] The announcement does not establish that SKY-001 is commercially available: the product is currently in research preview.[1]

Skyfall says SKY-001 can begin with text inputs or 2D concept sketches, then generate and explore multiple design iterations in parallel.[1] It models construction as step-by-step parametric operations, including 2D sketch strokes and 3D feature extrusions.[1] The company says the system assembles components into 3D objects for visualization in standard CAD software such as Autodesk Fusion, and is designed to maintain structural validity throughout the design tree.[1] These described capabilities give engineering teams specific workflow claims to evaluate during a preview, rather than evidence of production readiness.

On a reported 3D CAD benchmark, SKY-001 scored 79.7% IoU, compared with 79.1% for GPT-6 Astra; the reported run times were 3.9 seconds and 199 seconds, respectively.[3] A case was reported to cost $0.005 for SKY-001 and $0.75 for GPT-6 Astra.[3] For Skyfall’s BIM benchmark, the company reported 43.6% IoU for SKY-001, compared with 42.3% for the best-performing LLM, Gemini 3.8 Flash; SKY-001 took 1.22 seconds per case at about $0.0024 per case.[3] The reported 3D CAD benchmark results are self-reported, so teams should treat them as claims to validate against their own tasks.[3]

SKY-001 access is currently waitlist-only, and Skyfall is targeting commercial deployment for early 2027.[3] That timeline makes it relevant for teams deciding whether to monitor the preview now or plan around a later commercial release. Skyfall AI was founded by the team behind Maluuba, which Microsoft acquired, and its listed backers include Fidelity, Touring Capital, M13 and iNovia Capital.[3] Skyfall also says lower marginal compute costs will electrify adoption; that is the company’s view, not a reported adoption result.[1]