Analyzing Frontier Model Progress with My Favourite Game: Prince of Persia

A developer tested frontier AI models' ability to port the classic 1989 game Prince of Persia from assembly code to C# by providing only prompts and gameplay feedback. Claude Opus 5 significantly outperformed earlier models by diagnosing the original architecture problem, rebuilding the engine to match the original's frame-based animation system, and extracting game data directly from the DOS version—achieving a playable result that Claude Opus 4.6 and OpenAI Codex could not accomplish.