An Indian entrepreneur has introduced a specialized artificial intelligence model designed to solve complex engineering and design problems, claiming it can match or outperform leading AI models while costing significantly less to run.

Sam Pasupalak, co-founder and CEO of Skyfall AI, has unveiled SKY-001, an AI model developed for engineering applications involving computer-aided design (CAD) and building information modelling (BIM). According to the company’s published benchmarks, SKY-001 achieved results comparable to GPT-6 Astra on a specific CAD construction task while costing approximately 150 times less per sample.

The model is designed to help AI systems understand how physical objects and structures can be constructed, rather than simply generating text or images. It breaks complex designs into smaller components and organizes the steps required to build them.

The announcement highlights an important development in the AI industry: specialized models may be able to deliver strong results on particular technical tasks without requiring the computing costs associated with larger, general-purpose AI systems.

However, the reported performance figures come from Skyfall AI’s own tests and have not been independently verified. They should be understood as benchmark-specific claims rather than evidence that SKY-001 is better than GPT-6 Astra across all applications.

What Is SKY-001 AI?

SKY-001 is an engineering-focused AI model developed by Skyfall AI. The company describes it as an engineering world model intended to help users design and reason about physical objects before building them.

Unlike a conventional chatbot that primarily generates text, an engineering model needs to represent the relationships between parts, dimensions, structures and construction steps.

For example, designing a complex mechanical object may involve multiple assemblies, individual components, sketches and geometric constraints. A building design may require rooms, walls, floors and other elements to fit together within a larger structure.

SKY-001 aims to handle these challenges by breaking a large design problem into smaller, manageable stages.

According to the company, the model can take text or image inputs and generate preliminary three-dimensional design assets. These outputs can help users explore possible designs before refining them into more detailed engineering work.

The model is intended for applications involving:

• Computer-aided design (CAD)

• Building information modelling (BIM)

• Mechanical and product design

• Aerospace engineering

• Printed circuit board (PCB) design

• Other complex physical-design applications

These are potential application areas for engineering-focused AI. The availability and quality of specific workflows will depend on the model’s supported capabilities and further validation.

Who Is Sam Pasupalak?

Sam Pasupalak is an Indian entrepreneur originally from Bhubaneswar, Odisha, who is based in San Francisco, United States.

He is the co-founder and CEO of Skyfall AI, the company behind SKY-001.

Pasupalak studied mathematics and computer science at the University of Waterloo in Canada, completing his bachelor’s degree in 2011. He later earned a master’s degree in political science and government from the London School of Economics in 2019.

Before launching Skyfall AI in 2025, he co-founded Maluuba, an artificial intelligence startup established in 2011. He served as its CEO, and Microsoft acquired the company in 2017.

His experience in AI research and entrepreneurship forms part of the background to Skyfall AI’s work on specialized models for engineering and physical design.

The company says its team spent approximately 18 months developing SKY-001 before announcing the model and sharing benchmark results.

SKY-001 vs GPT-6 Astra: What Do the Benchmark Results Show?

One of the most attention-grabbing claims surrounding SKY-001 is its reported cost advantage over GPT-6 Astra.

In the CAD construction benchmark published by Skyfall AI, the company compared its model with several leading AI systems using measures of construction quality, speed and cost.

Here is how SKY-001 compared with GPT-6 Astra in the company’s reported results.

Source: Skyfall AI’s published benchmark. Results are company-reported and have not been independently verified.

1. SKY-001 Reportedly Costs 150 Times Less

The company reports that one CAD benchmark sample costs approximately $0.005 to process with SKY-001, compared with $0.75 for GPT-6 Astra.

The calculation is straightforward:

• GPT-6 Astra: $0.75 per sample

• SKY-001: $0.005 per sample

• Reported difference: 150 times

At those stated rates, processing 1,000 comparable samples would cost approximately $5 with SKY-001 and $750 with GPT-6 Astra.

This is a substantial difference if the two systems are being compared under equivalent conditions and produce results of comparable usefulness.

However, the calculation reflects the company’s stated benchmark costs. It does not automatically represent the total cost of deploying either system in production, where engineering work, infrastructure, integration and quality assurance can affect the overall expense.

2. SKY-001 Scores Slightly Higher on the Reported CAD Metric

Skyfall AI reports that SKY-001 achieved 79.7% on the Intersection over Union (IoU) metric, compared with 79.1% for GPT-6 Astra.

IoU measures the overlap between a predicted result and a reference result. In this benchmark, it is used to assess how closely the model’s generated CAD construction aligns with the reference design.

The difference is 0.6 percentage points.

This suggests that, on the particular task tested, the company’s model achieved a slightly higher score while using substantially less reported cost per sample.

However, a small benchmark difference does not necessarily establish a meaningful overall advantage. The result would need to be evaluated across more examples, different design types and independent tests before drawing broader conclusions.

3. The Company Reports Major Speed Improvements

Skyfall AI also claims that SKY-001 is between 51 and 125 times faster than the frontier models included in its comparisons.

The exact speed advantage depends on which model and task are being compared.

Faster output can be particularly useful for engineering workflows that involve generating many design alternatives, evaluating intermediate results or repeatedly adjusting components.

Nevertheless, the reported speed figures should be interpreted in the context of the company’s benchmark methodology. Independent testing would help establish how the model performs under different workloads and deployment conditions.

How Does SKY-001 Work?

SKY-001 uses a staged approach to complex engineering problems.

Rather than attempting to produce an entire design in one large generation step, the model organizes the task into smaller construction stages. This helps it manage the relationships between individual components and the final object.

Step 1: Understand the Design Request

The process begins with an input such as a text description, a visual reference or a basic outline.

For example, a user might describe a physical product or provide an image representing the intended design.

The model then uses that input to establish a preliminary representation of the object.

Step 2: Break the Object Into Components

Complex designs often contain multiple parts that must fit together correctly.

SKY-001 is designed to divide a project into assemblies and individual components. For building-related tasks, a similar approach can be used to break a building outline into spaces and rooms.

This makes the overall design problem easier to organize than treating the entire structure as one indivisible output.

Step 3: Plan the Construction Sequence

The model then plans intermediate design states and the steps needed to move between them.

According to Skyfall AI, this hierarchical approach helps the system handle projects that may involve hundreds or thousands of individual steps.

The idea is to maintain consistency between the smaller components and the larger design as the construction process progresses.

Step 4: Generate Preliminary 3D Assets

The system can produce preliminary three-dimensional design assets that users can examine and refine.

These assets can help engineers explore different directions before committing to detailed modelling.

They should not automatically be treated as manufacturing-ready engineering files. Professional use may require CAD validation, geometric checks, engineering analysis and review by qualified specialists.

What Is an Engineering World Model?

A world model is an AI system designed to represent aspects of how the world works and how objects or environments relate to one another.

In physical-design applications, this can include spatial relationships, object geometry and the consequences of changing parts of a structure.

A general-purpose language model can explain how a mechanical device works or help write code for a design tool. However, producing a valid physical design may require additional capabilities beyond generating a convincing explanation.

An engineering-focused world model aims to represent the design process more directly.

For example, a system may need to understand that a component must fit within an assembly, that a building room must fit within a floor plan or that modifying one part can affect other components.

SKY-001 is designed around this type of structured, multi-stage engineering workflow.

The broader idea is that specialized AI systems may be able to perform particular technical tasks more efficiently than models designed to handle a wide variety of unrelated requests.

Why Could Specialized AI Models Be Cheaper?

Large general-purpose AI models are built to support a broad range of activities, including writing, coding, research, reasoning and professional work.

That versatility can be valuable, but not every engineering task requires the full range of capabilities offered by a general-purpose system.

A specialized model can instead focus on a narrower set of problems and use an architecture and workflow designed for those tasks.

There are several reasons this approach could reduce costs.

Focused optimization: The model can be optimized for particular engineering operations rather than a wide variety of tasks.

Structured processing: Breaking a complex problem into smaller stages may make individual steps easier to manage and evaluate.

Reduced unnecessary computation: A specialized workflow may avoid spending resources on capabilities that are not needed for a particular task.

Task-specific outputs: The system can concentrate on producing useful engineering representations instead of general conversational responses.

These principles can make specialized AI attractive for repetitive technical workflows. However, the actual savings depend on model design, hardware, inference costs, workload complexity and the quality of the output.

How Could SKY-001 Be Used in Real Engineering Projects?

If the model proves reliable beyond its published benchmarks, it could help support several types of engineering work.

Product and Mechanical Design

Engineers often explore multiple versions of a product before selecting a final design.

An AI model that can generate preliminary three-dimensional components and organize construction steps could help teams evaluate alternatives more quickly.

Human engineers would still need to check dimensions, tolerances, materials and manufacturing requirements.

Building Design and BIM

Building information modelling involves creating and managing digital representations of buildings and their components.

SKY-001’s reported ability to work with building outlines and divide them into individual spaces could be useful in preliminary layout generation.

However, building designs must satisfy applicable structural, accessibility, fire-safety and other regulatory requirements. AI-generated layouts should not be treated as certified construction plans without professional review.

Aerospace Engineering

Aerospace design involves complex geometry, strict tolerances and demanding safety requirements.

Specialized AI could potentially assist with early-stage exploration, component generation or repetitive modelling tasks.

Any application in safety-critical aerospace systems would require extensive validation and compliance with relevant engineering standards.

PCB and Electronic Design

Printed circuit board design requires careful coordination of components, electrical connections and physical layout.

AI tools could potentially help engineers explore layouts and organize design alternatives, although practical use requires verification of electrical constraints, manufacturing rules and signal integrity.

Scientific and Molecular Design

Some specialized design systems can be used to explore molecular structures and other scientific representations.

Although Skyfall AI identifies molecular design among the areas relevant to its broader ambitions, the company’s CAD benchmark alone does not establish performance in molecular modelling or scientific discovery.

Each application would require its own appropriate evaluation.

Can SKY-001 Replace GPT-6 Astra?

Not on the evidence currently available.

SKY-001 is designed for specialized engineering workflows, while GPT-6 Astra is a general-purpose AI model intended to support a broad range of complex tasks.

The comparison therefore depends on what the user needs to accomplish.

For a particular CAD construction task, SKY-001 may offer an attractive combination of reported quality, speed and cost. For software engineering, research, document analysis, general reasoning or tasks outside its supported design capabilities, the CAD results do not establish that it is the better choice.

A fair comparison would need to evaluate both models on equivalent inputs, consistent hardware and cost assumptions, clearly defined output requirements and independent quality checks.

The most useful question is not simply which model has the lowest cost. It is which system delivers the required result accurately, reliably and efficiently for a specific task.

What Does This Mean for the Future of AI?

The SKY-001 announcement reflects a wider shift in AI development toward specialized systems.

As businesses adopt AI for technical work, they increasingly need models that can deliver predictable results at an acceptable cost. A model that performs well on one narrowly defined task may be more economical than using a general-purpose system for every step.

This does not mean general-purpose AI models are becoming unnecessary. Broad models remain useful when a workflow requires different kinds of reasoning, communication and problem-solving.

Instead, the industry may move toward a combination of general-purpose assistants and specialized models that handle particular parts of a larger workflow.

For example, a general-purpose AI system might help define a product’s requirements, while an engineering model generates preliminary geometry. Specialized tools could then validate dimensions, test performance and prepare the design for professional review.

If these systems can work together reliably, engineering teams may gain new ways to reduce repetitive work and explore design options.

Frequently Asked Questions

1. What is SKY-001 AI?

SKY-001 is an engineering-focused AI model developed by Skyfall AI. It is designed to assist with complex physical-design tasks involving CAD, BIM and preliminary three-dimensional assets.

2. Who developed SKY-001?

The model was developed by Skyfall AI, co-founded by Indian entrepreneur Sam Pasupalak, who serves as the company’s CEO.

3. Is SKY-001 better than GPT-6 Astra?

Skyfall AI reports that SKY-001 slightly outperformed GPT-6 Astra on a specific CAD construction benchmark. This does not establish that it is better across general-purpose AI tasks or every engineering application.

4. Why is SKY-001 claimed to be 150 times cheaper?

The company’s benchmark reports a cost of $0.005 per sample for SKY-001, compared with $0.75 for GPT-6 Astra. That is a 150-fold difference under the reported comparison, but real-world costs can vary.

5. What does the 79.7% IoU score mean?

IoU, or Intersection over Union, measures the overlap between a predicted result and a reference result. Skyfall AI reports a score of 79.7% for SKY-001 and 79.1% for GPT-6 Astra on its CAD construction benchmark.

6. Can SKY-001 generate 3D designs?

Skyfall AI says the model can turn text or image inputs into preliminary 3D design assets. The usefulness of those outputs for professional engineering depends on the task and the validation required.

7. Is SKY-001 available to everyone?

The company’s announcement and benchmark describe the model’s capabilities, but availability and access conditions should be checked through Skyfall AI’s official website.

Conclusion

SKY-001 is an interesting example of how specialized AI models could change technical design workflows. Developed by Skyfall AI under Indian entrepreneur Sam Pasupalak, the model focuses on engineering tasks that require structured planning, component relationships and preliminary physical-design generation.

The company’s published results report CAD construction quality of 79.7% IoU at a cost of $0.005 per sample, compared with 79.1% and $0.75 for GPT-6 Astra. If independently validated under comparable conditions, these results could demonstrate a substantial cost advantage for the specific task tested.

However, the claims remain benchmark-specific and company-reported. Broader testing will be necessary to establish the model’s reliability across different engineering applications.

The larger takeaway is that AI competition is no longer only about building the biggest general-purpose model. Specialized systems that solve particular problems efficiently could also play an important role in the future of engineering, product design and industrial automation.