One prompt, 8,429 wrong pixels to 2: how Claude Opus 5.5 finally made Prince of Persia look right
Analyzing Frontier Model Progress with My Favourite Game: Prince of Persia
I handed the original Apple II Prince of Persia source to each new frontier model and asked it to port the game to C#. Opus 4.6 picked the wrong architecture; Codex polished the surface; Opus 5, given tools to watch the real game, rebuilt the engine and produced a playable version. Opus 5.5, from a single prompt, ported SDLPoP's room-drawing routine and cut the first screen's pixel differences from 8,429 to 2.
The biggest jump wasn't only the models getting smarter. It came from giving them a way to see the original and test themselves against it.