Skyfall AI’s SKY-001 Claims to Match GPT-6 Astra on Engineering Tasks at a Fraction of the Cost
Indian entrepreneur Sam Pasupalak, co-founder and chief executive officer of Skyfall AI, has introduced SKY-001, an artificial intelligence model designed specifically for engineering and design applications. The company claims that its model can deliver results comparable to GPT-6 Astra on selected computer-aided design benchmarks while completing tasks considerably faster and at a much lower cost.
The announcement highlights a specialised approach to artificial intelligence that focuses on visual reasoning, spatial relationships and complex engineering workflows rather than relying solely on conventional language-based generation. However, the performance figures have been published by Skyfall AI itself and have not been independently verified.
What Is SKY-001?
SKY-001 is described by Skyfall AI as an engineering world model designed to create and organise complex physical designs. According to the company, the system can work with computer-aided design, commonly known as CAD, and building information modelling, or BIM.
CAD software is used to create detailed product designs, mechanical components and three-dimensional models. BIM technology is commonly used in architecture and construction to organise building layouts, rooms and related structural information.
Skyfall AI says existing language-based AI systems can struggle with engineering tasks that require many connected components to remain geometrically consistent. A design may look convincing but still contain structural relationships or manufacturing steps that are difficult to implement in the real world.
SKY-001 aims to address this challenge by breaking a complex design into smaller sections. Instead of generating an entire object in one continuous process, the model organises the design into assemblies, individual parts and the steps needed to construct them.
According to the company’s technical explanation, this hierarchical process is intended to help the system maintain consistency across long sequences of design decisions.
SKY-001 Versus GPT-6 Astra: What the Company Reports
Skyfall AI published benchmark comparisons between SKY-001 and several advanced AI models, including GPT-6 Astra. The results focus on the quality of generated designs, the time required to complete a task and the estimated cost of processing each case.
In its computer-aided design construction benchmark, Skyfall AI reports that SKY-001 achieved an Intersection over Union score of 79.7 percent. GPT-6 Astra recorded 79.1 percent on the same benchmark, according to the company’s published results.
Intersection over Union, or IoU, is a measurement used to assess how closely a generated shape overlaps with a reference shape. A higher score generally indicates a closer match between the generated result and the reference design.
The company also reports a substantial difference in processing time. SKY-001 completed the tested CAD construction task in approximately 3.9 seconds per case, while GPT-6 Astra took around 199 seconds in the comparison.
The reported cost difference was also significant. Skyfall AI estimates the cost of running SKY-001 at 0.005 US dollars per CAD case, compared with 0.75 US dollars for GPT-6 Astra.
Based on these figures, the company describes SKY-001 as approximately 51 times faster and 150 times cheaper for the particular CAD task evaluated.
These figures should be understood within the limits of the reported test. They do not establish that SKY-001 is faster, cheaper or more accurate than GPT-6 Astra for every engineering application or general AI task.
How the Model Approaches Engineering Design
Skyfall AI says SKY-001 uses a hierarchical design process. It first establishes the overall structure of an object and then divides that structure into smaller components. The system subsequently generates visual construction steps for individual parts and combines them into a larger design plan.
For example, a complex mechanical product may contain multiple assemblies, each made up of smaller components. The company’s approach is intended to keep the relationships between those components consistent as the design develops.
The model is also designed for building-related tasks. In BIM applications, the system can reportedly start with a building outline and organise it into layouts, individual units and rooms.
This method is intended to make lengthy design tasks more manageable by separating them into stages. Skyfall AI argues that such a process is better suited to engineering problems involving geometry, spatial constraints and sequences of construction operations.
The practical value of this approach will depend on how reliably the model handles different design requirements, whether its output can be used in established engineering workflows and how it performs on tasks outside the company’s published tests.
Reported Results for Building Design
Skyfall AI has also published results for building information modelling tasks. According to the company, SKY-001 achieved an IoU score of 43.6 percent on its BIM construction benchmark, compared with 40.6 percent for GPT-6 Astra.
The company reports that SKY-001 completed a BIM case in approximately 1.22 seconds, while GPT-6 Astra required around 144.6 seconds in the comparison. The estimated cost was 0.0024 US dollars per case for SKY-001 and 0.446 US dollars for GPT-6 Astra.
These figures suggest that the model could be useful for specialised workflows where repeated design generation and processing costs matter. However, the scores and cost estimates reflect the company’s own evaluation setup. Independent testing would be needed to establish how the model performs under different conditions and against other engineering tools.
Who Is Sam Pasupalak?
Sam Pasupalak is the co-founder and CEO of Skyfall AI, the company developing SKY-001. The company’s announcement presents the model as part of its broader effort to apply world-model approaches to engineering and the design of physical systems.
Skyfall AI says its research aims to extend beyond conventional text generation towards systems that can represent complex structures and plan how those structures are built. Its longer-term ambitions include applications across additional technical and industrial fields, although these broader possibilities should not be confused with capabilities that have already been independently demonstrated.
What the Claims Mean for the AI Industry
The SKY-001 announcement illustrates the growing interest in specialised AI systems built for particular professional tasks. General-purpose AI models are designed to support a wide range of activities, while specialised systems may be optimised for narrower workflows such as mechanical design or building layouts.
If Skyfall AI’s reported performance can be reproduced independently, a model that produces comparable engineering results at a fraction of the time and cost could attract interest from designers, engineering teams and construction-related businesses. Lower processing costs could be especially relevant for workflows that require many design iterations.
Nevertheless, benchmark results alone do not establish commercial readiness. Important questions remain about reliability, the quality of outputs in real projects, compatibility with existing software, the handling of complex constraints and the costs associated with integrating the model into professional workflows.
As of the company’s published announcement, SKY-001 is being presented through its website and a waitlist rather than as a broadly available model with a public price list.
Conclusion
Skyfall AI’s SKY-001 is an emerging engineering-focused AI model that the company says can match GPT-6 Astra on selected CAD and BIM benchmarks while requiring substantially less processing time and money. Its hierarchical approach to design generation offers a different direction from conventional language-based AI systems.
For now, the headline performance figures should be treated as company-reported claims rather than independently established results. External testing, wider evaluations and real-world use will be important in determining whether SKY-001 can deliver the same advantages across practical engineering projects.
