An Indian AI startup is claiming that its latest model can generate engineering designs faster and at a fraction of the cost of leading AI systems, including GPT-6 Astra. The model, called SKY-001, has been developed by Skyfall AI for computer-aided design (CAD), architectural planning and other engineering applications.
Sam Pasupalak, co-founder and CEO of Skyfall AI, announced the model in a post on X, outlining its reported performance against other AI systems tested by the company. One of the most notable figures in the comparison is the reported cost of generating a single CAD sample: $0.005 for SKY-001, compared with $0.75 for GPT-6 Astra.
The company also claims that its model delivers slightly higher accuracy while running considerably faster and costing less. However, the performance figures come from Skyfall AI’s own tests and have not been independently verified.
What Is SKY-001 and How Does It Work?
SKY-001 is designed to handle spatial reasoning and engineering design tasks, rather than focusing exclusively on text generation or code-based outputs.
According to Skyfall AI, the model functions as a spatial world model that plans engineering and architectural designs directly in the visual world. Its intended applications include computer-aided design (CAD), which is used to create detailed designs of products, machinery and structures, and building information modelling (BIM), which produces digital representations of buildings and their construction-related information.
The system uses a two-stage architecture to break down complex design tasks into smaller, more manageable steps.
The first stage, called the Hierarchical Composer, develops the overall design and divides it into smaller assemblies and individual components. It then refines these parts by filling in missing geometry, smoothing surfaces and adding finer details.
The second stage, known as the Construction World Model, determines how the components can be constructed step by step.
This approach is intended to help the model move from an overall design concept to more detailed components and construction sequences. Rather than treating an engineering task as a single output, SKY-001 breaks the process into stages covering both design and construction.
SKY-001 vs GPT-6 Astra: What the Numbers Show
Skyfall AI has shared several performance claims comparing SKY-001 with leading AI models, including GPT-6 Astra. The company says its system offers advantages in speed, cost and CAD accuracy.
According to the figures shared by Pasupalak, SKY-001 is between 51 and 125 times faster than the leading models included in the company’s tests. Its reported cost is also 150 to 210 times lower, depending on the model used for comparison.
The specific CAD comparison with GPT-6 Astra highlights the difference in operating costs and the relatively narrow gap in reported accuracy.
| Performance metric | SKY-001 | GPT-6 Astra |
|---|---|---|
| Cost per CAD sample | $0.005 | $0.75 |
| Intersection over Union (IoU) score | 79.7% | 79.1% |
The reported cost difference is substantial. At $0.005 per sample, SKY-001 costs less than one cent to generate a CAD sample, compared with 75 cents for GPT-6 Astra.
The accuracy results show a smaller difference. SKY-001 recorded an IoU score of 79.7%, against 79.1% for GPT-6 Astra, giving the Indian model a lead of 0.6 percentage points.
Intersection over Union, or IoU, measures the overlap between a predicted result and a reference result. In design-related evaluations, it can be used to assess how closely a generated shape or geometry matches the expected output. A higher score indicates greater overlap under the metric being used.
While the reported results favour SKY-001, the difference in accuracy is modest compared with the claimed differences in speed and cost.
Why the Cost Difference Matters for Engineering AI
The economics of generating engineering designs could be an important consideration for businesses exploring AI-assisted design workflows.
CAD and BIM applications involve detailed representations of physical objects and buildings. If a model can generate usable designs at a lower cost, it could make repeated design iterations more economical for teams working on complex projects.
Skyfall AI’s reported pricing comparison points to this possibility, particularly for workloads that require large numbers of generated samples. However, the figures alone do not establish how the model performs across different engineering tasks or whether its outputs are ready for use in professional design environments.
The usefulness of such a system also depends on the quality of its generated geometry, its ability to follow engineering constraints and the extent of human review required before a design can be used.
Independent Verification Remains the Key Question
Despite the performance claims, SKY-001’s reported advantage should be viewed in the context of the testing conducted by its developer.
Skyfall AI has presented its own comparisons, but the results have not been independently verified. The available figures therefore do not establish whether the model consistently outperforms GPT-6 Astra or other leading systems across a broader range of engineering applications.
The comparison also leaves questions about how the models were evaluated, including the tasks used, the testing conditions and whether the same standards were applied to every system.
For now, SKY-001 presents an interesting approach to AI-assisted engineering, with its focus on spatial design, component-level refinement and construction planning. Its reported cost advantage is particularly striking, while the accuracy comparison suggests a much narrower lead.
Independent testing and further evidence of performance across real-world design tasks will be important in determining whether Skyfall AI’s claims translate into a practical advantage for engineers and architects.
Also Read: Know About Jio IPO: Price Band, GMP, Dates and Key Details
