Claude Opus 5.5 Max Builds Game After Three Hours

Technologist Simon Willison recently demonstrated the expanding creative capabilities of advanced AI models by prompting Anthropic's Claude Opus 5.5 Max to construct a complete web-based application. Using the standard Claude iPhone application in standard chat mode—rather than relying on developer-focused environments such as Claude Code mode—Willison tasked the system with creating a Three.js application titled 'Tree appreciator.' The original prompt instructed the AI to present users with two algorithmically generated trees and build gameplay mechanics centered around appreciating their differences, explicitly asking the model to "think hard about how to make sure it is both fun and beautiful."

In response, the model executed an extensive generation cycle, crunching for three hours and seven minutes before delivering the completed project. The result was a surprisingly complex, cozy spot-the-difference game equipped with custom background music, real-time animation, and an interactive achievement system. Willison observed that including the simple directive to ensure the game was fun worked uncomfortably well, allowing the system to handle game design, aesthetic tuning, and feature implementation concurrently.

The successful generation of a multifaceted game from a single prompt illustrates the growing utility of giving large language models extensive processing time for end-to-end tasks. For developers and technical practitioners, this experiment demonstrates that consumer-level mobile interfaces can now support extended, autonomous code generation cycles. As frontier models increasingly handle holistic asset integration, audio synthesis, and logic design without manual oversight, developers gain powerful capabilities for rapid prototyping and automated software generation directly from loose conceptual briefs.