James Ding
Jun 17, 2026 22:43
From 3D reconstructions to synthetic cities, Claude Opus 4.8 helped developers build cutting-edge tools during a one-day hackathon in San Francisco.
Over 300 developers gathered in San Francisco on May 13, 2026, to push the limits of Anthropic’s Claude Opus 4.8 during a 12-hour hackathon. The competition, which had 1,500 applicants and welcomed 310 participants, provided $500 in credits to each team to transform ideas into fully functional demos. The event showcased the versatility of Claude Opus 4.8, Anthropic’s flagship AI model, which launched just two weeks prior on May 28.
The winning projects spanned cultural preservation, urban simulation, and robotics training—each leveraging Opus 4.8’s dynamic workflows and massive 1-million-token context window to tackle complex, multi-step tasks. Here’s what set the top three apart:
First Place: Tekton
Developed by: Holly Tang and Austin Burgess
Tekton is a 3D reconstruction platform designed to digitally preserve historical architecture. By integrating schematics, photographs, and records, it creates highly accurate models of buildings like Notre-Dame’s spire or Tang Dynasty temples. Each component is traceable to its historical source, with Opus 4.8 verifying all details through independent sub-agents and self-correcting loops to ensure accuracy.
The project has broad potential applications, including cultural preservation, academic research, and restoration planning. Tang and Burgess aim to open-source Tekton, inviting museums and governments to contribute. According to Burgess, their meticulous project planning, mapped out via a 50-task Notion board, was key to executing the complex build in just one day.
Second Place: Sim Francisco
Developed by: Tanmayi Priya Dasari and Tejas Prabhune
This project created a digital twin of San Francisco, populated with 10,000 synthetic residents modeled on U.S. Census data. The tool polls this simulated population to predict real-world outcomes, such as election results and ballot measures, with high accuracy. For example, it forecast March 2024’s Prop A vote to within 0.4% of the actual result.
Opus 4.8 not only handled the backend but also optimized the system by clustering 10,000 residents into 300 representative personas, reducing costs without sacrificing accuracy. The creators see applications beyond political forecasting, including urban planning and market research.
Third Place: Custom Universe
Developed by: Jake Stevens and Mauricio Pereira
Custom Universe transforms a single photo into a photorealistic 3D scene that can be edited in real time. Aimed at robotics labs, it generates synthetic training data for robots operating in specific environments. The tool relies on Claude Opus 4.8 to build its pipeline, allowing scanned objects to be incorporated seamlessly. The developers, who met during the hackathon, plan to keep the tool free and open-source for widespread adoption.
Why It Matters
The hackathon demonstrated the power of Claude Opus 4.8’s enhanced capabilities. Released in late May as a bridge to Anthropic’s forthcoming Mythos systems, the model’s features—dynamic workflows, improved reasoning, and cost-efficient modes—enable developers to tackle large-scale, multi-step challenges in new ways. These projects underscore its potential for solving real-world problems, from cultural preservation to urban forecasting and robotics training.
With official support for Opus 4.8 slated through May 2027, the model remains Anthropic’s cornerstone for high-risk and resource-intensive tasks, complementing the newly launched Fable 5 system. As developers continue to explore its possibilities, Opus 4.8 could shape the next wave of AI-driven applications.
Image source: Shutterstock
