Indian entrepreneur Sam Pasupalak has introduced SKY-001, an AI model for engineering design that reportedly matches GPT-6 Astra’s accuracy at a significantly lower cost.
Indian entrepreneur Sam Pasupalak, co-founder and CEO of Skyfall AI, has introduced SKY-001, an artificial intelligence model designed to handle complex engineering and design tasks. The company claims its model can deliver comparable or better results than leading AI systems while operating at a substantially lower cost.
Developed over 18 months, SKY-001 focuses on computer-aided design (CAD) and building information modelling (BIM). Pasupalak announced the model on social media platform X, sharing benchmark results comparing its performance with other AI systems, including GPT-6 Astra.
The reported findings suggest that specialised AI models could offer a more economical approach to engineering workflows. However, the results are based on the company's own tests and have not been independently verified.
Many AI systems can generate designs and write code for complex products, but producing a complete, usable engineering design involves more than creating an initial output. Designs must account for individual components, their relationships and the steps required to assemble them correctly.
SKY-001 aims to address this challenge through a structured, multi-stage design process rather than generating an entire project in one step.
For a complex product, the model first divides the overall design into assemblies and individual components. It then determines how those parts should be constructed, using visual representations, sketches and sequential building instructions.
This method is intended to help the system maintain consistency between individual parts and the final design.
For building information modelling tasks, SKY-001 can begin with a building outline and organise the structure into separate units and rooms.
According to Skyfall AI, breaking projects into smaller stages enables the model to manage engineering workflows involving hundreds or thousands of steps while preserving the relationships between components.
Skyfall AI founder Sam Pasupalak claims that SKY-001 delivers substantial improvements in processing speed and cost compared with the frontier models evaluated by his team.
According to the company's reported comparisons, SKY-001 was between 51 and 125 times faster and approximately 150 to 210 times cheaper, depending on the model and task being evaluated.
The CAD benchmark shared by Pasupalak provides a specific example of the reported cost difference.
|
Performance metric |
SKY-001 |
GPT-6 Astra |
|---|---|---|
|
Cost per CAD sample |
$0.005 |
$0.75 |
|
Intersection over Union (IoU) score |
79.7% |
79.1% |
Based on these figures, the reported cost of running the CAD task on GPT-6 Astra was 150 times higher than on SKY-001.
The accuracy results showed a relatively small difference. SKY-001 achieved an IoU score of 79.7%, compared with 79.1% for GPT-6 Astra.
IoU, or Intersection over Union, measures how closely a generated result overlaps with a reference result. In design-related evaluations, it can help assess the similarity between an AI-generated output and the expected design.
However, these figures should be interpreted carefully. They represent Skyfall AI's reported results for specific tasks and do not establish that SKY-001 is superior to general-purpose AI models across all applications. Independent testing would be needed to validate the performance claims.
Engineering design and construction projects often involve complicated workflows, numerous components and repeated revisions. AI systems capable of handling these processes could help reduce the time and expense associated with developing and refining designs.
By focusing on CAD and BIM, SKY-001 targets tasks relevant to product engineering, architectural design and building planning.
Its staged approach could be particularly useful for projects where individual components must fit together correctly and remain consistent with the larger structure.
The reported reduction in processing costs could also make repeated design iterations more economical. Nevertheless, practical benefits will depend on how the model performs in real-world projects, including its reliability, integration with existing software and ability to meet professional engineering requirements.
Sam Pasupalak is an entrepreneur originally from Bhubaneswar, Odisha, who is now based in San Francisco, United States. He is the co-founder and chief executive officer of Skyfall AI, the company developing SKY-001.
Pasupalak earned a bachelor's degree in mathematics and computer science from the University of Waterloo in Canada in 2011. He later completed a master's degree in political science and government at the London School of Economics in 2019. His studies at Waterloo included areas such as artificial intelligence, robotics and human-computer interaction.
Before launching Skyfall AI in 2025, Pasupalak co-founded Maluuba in 2011 and served as its CEO. The AI research startup focused on artificial intelligence and language understanding and was acquired by Microsoft in 2017.
At Maluuba, he led teams of researchers and engineers working on AI technologies, experience that preceded his work on specialised engineering models.
Conclusion
Skyfall AI's SKY-001 highlights an emerging approach to artificial intelligence focused on specialised, multi-step engineering tasks. Its reported benchmarks suggest that the model could deliver comparable CAD accuracy to GPT-6 Astra at a substantially lower cost. However, the claims remain company-reported, and independent evaluations will be important in determining how effectively SKY-001 performs across practical engineering and construction applications.