On July 10, 2026, Anthropic released a video demonstrating how the Claude 3.5 Sonnet model designed and supervised the construction of a 1:1000-scale miniature New York City. This isn’t just a showcase of AI capabilities—it’s a glimpse into a new era of spatial design, where algorithms become co-creators of physical structures.
July 2026 marked another breakthrough in AI applications. This time, it wasn’t about generating text or images—it was about something far more tangible: a physical city model. The video Building a Miniature New York with Claude, published on July 10 on Anthropic’s YouTube channel, shows how the latest version of the model—Claude 3.5 Sonnet—not only designed but also supervised the construction of a 1:1000-scale miniature New York. This project bridges the digital and physical worlds, with implications far beyond an impressive demo.
Claude 3.5 Sonnet: A New Era in 3D Design
The release of Claude 3.5 Sonnet on June 20, 2026 introduced several key improvements essential for this project. Compared to earlier versions, the new model offers:
- Extended context window (200K tokens) – enabling processing of vast datasets like maps or 3D models.
- Enhanced CAD and 3D printing integration – Claude can generate STL files required for 3D printing and collaborate with design software like Blender or AutoCAD.
- Physics simulations – the model can predict structural stability, a critical factor in building a miniature city.
Notably, Claude 3.5 Sonnet doesn’t operate in isolation. The project leveraged OpenStreetMap and NYC OpenData as foundational datasets for the digital city model. This underscores the importance of AI integration with existing data sources—without them, the project wouldn’t have been feasible.
For a deeper dive into how Claude’s "mind" works and its limitations, check out my previous post: What’s Really Inside Claude’s "Mind"?.
How the Miniature New York Was Built
The model’s creation process can be broken down into several key phases, each highlighting different capabilities of Claude 3.5 Sonnet.
1. Data Generation and Preliminary Design
The first step involved collecting and processing New York City data. Claude utilized:
- OpenStreetMap – for street layouts, buildings, and green spaces.
- NYC OpenData – for building heights, facade materials, and architectural details.
- Satellite imagery – to incorporate the city’s real-world appearance.
Based on this data, Claude generated vectorized 3D models of city elements like skyscrapers, parks, and bridges. Crucially, the model had to account for technical constraints—such as maximum 3D-printable wall angles or minimum wall thickness for structural integrity.
2. Project Optimization
Claude didn’t settle for a single design variant. Instead, it generated multiple project iterations, varying in:
- Detail scale (e.g., facade intricacy levels).
- Color schemes (e.g., uniform tones vs. realistic facade colors).
- Infrastructure layouts (e.g., street, sidewalk, and green space placements).
Next came physics simulations. Claude tested whether skyscrapers would collapse under their own weight, if bridges could handle load stress, and how the model would appear under different lighting conditions. This demonstrates AI’s ability not just to design but also to test and refine its concepts.
3. 3D Printing and Assembly
Once Anthropic’s team approved the design, it was time for physical realization. The model was 3D-printed using Formlabs Form 4 printers, known for precision and speed. Materials included Standard Resin (for decorative elements) and Engineering Resin (for load-bearing parts).
Assembly was handled by Universal Robots’ UR5e robots. Claude generated step-by-step instructions for the robots, which assembled the model’s components according to the design. Another example of AI collaborating with robotics to tackle complex tasks.
4. Final Touches
The last phase added details to enhance realism. While most of the process was automated, some elements—like street painting or LED lighting—required human intervention. The lighting was integrated with a Raspberry Pi microcontroller, enabling control via simple software.
A notable detail: the entire process—from concept to final model—took about 5–6 weeks. An impressive timeline given the project’s scale and detail level.
Is Claude 3.5 Sonnet the Future of Design?
The miniature New York project proves AI can be more than a content-generation tool—it can be a co-creator of physical objects. Of course, AI still has gaps where human expertise shines, particularly in aesthetics and creativity. As IEEE Spectrum noted, some buildings in the model appear overly schematic, suggesting AI lacks the "human touch" in design.
Yet the potential is vast. Imagine AI designing not just city models but real buildings, bridges, or entire districts. We’re already seeing early AI applications in architecture—from optimizing room layouts to generating interior designs. Anthropic’s project is another step in this direction.
If you’re curious about how AI is transforming other fields, check out my post on how AI could save global ecosystems.
Challenges and Limitations
Despite its impressive results, the miniature New York project also highlights AI’s limitations:
- Data dependency – Claude couldn’t have built this model without OpenStreetMap or NYC OpenData. It underscores the critical role of high-quality, accessible datasets.
- Lack of "human" creativity – while AI can generate designs, it struggles to create something truly novel without a data-driven foundation.
- Cost and scalability – though groundbreaking, the project required advanced hardware and significant funding (estimated at $30,000–$50,000), limiting accessibility for smaller firms or individuals.
What’s Next?
The miniature New York project is just the beginning. AI’s role in designing and constructing physical objects is poised to grow. Here are a few potential directions:
- Designing future cities – AI could optimize street layouts, parks, and buildings for sustainability.
- Personalized products – from furniture to homes, AI could generate designs tailored to individual needs.
- Automated construction – combined with robotics, AI could oversee home or infrastructure builds, cutting costs and timelines.
Of course, many questions remain unanswered. Can AI generate truly original designs without relying on existing data? Will it account for cultural and social nuances in its projects? These topics will likely dominate discussions in the coming years.
For more on how AI is reshaping industries, see my post about Gemini 1.5 Pro and its million-token context window.
Summary
The Building a Miniature New York with Claude project isn’t just a flashy AI demo—it’s a harbinger of a new era in design and construction. Claude 3.5 Sonnet proved AI can be more than a content generator; it can be a co-creator of physical structures. While gaps remain—particularly in creativity and aesthetics—its capabilities are advancing at a breathtaking pace.
What we’ve seen here is only the beginning. In the future, AI may play an even larger role in designing cities, buildings, and infrastructure. Are we ready for such a future? It’s a question each of us will need to answer.
Comments