Four-word prompt makes GPT-6 Astra and Opus 5.5 each build a 3D runner game
Given the same four-word prompt "SUBWAY SURFER!!!!", both models generated a playable 3D runner in real time, which the author uses to compare taste and ability to fill in gaps beyond code quality.
The entire prompt was four words: SUBWAY SURFER!!!! Both models built a fully playable 3D runner from that alone. Side by side, you can watch which one actually nailed the vibe. GPT-6 Astra versus Opus 5.5, same one-line prompt, both on high reasoning mode, both generating a working game in real time. No spec sheet, no asset list, no reference screenshots. Just four words and an exclamation mark, and both models had to infer everything else. Track layout, hoverboard power-ups, coin collection, score counter, camera angle chasing the character down the rails. Watch how differently they interpreted the same four words. One leans into a cleaner, more literal subway aesthetic with working power-up timers. The other goes bolder with the color palette and graffiti detail, closer to the actual game's visual chaos. This is the part that's easy to miss if you only look at benchmark charts. Code quality is one thing. Taste, the ability to fill in everything a vague prompt left unsaid in a way that actually feels right, is a completely different skill, and it's the one that's hardest to put a number on. Neither model was told what Subway Surfers looks like. They both just knew. Which recreation would you actually want to keep playing?
This post quotes another post: x.com/h100envy/status/2109288230512394606