1 prompt. 38 model/effort. 38 variations of 1 game.

Making a game takes time, money, tokens. But what makes a game good? Was it fun to play? difficult enough that points feel earned, not handed? immersive enough to escape?

Play the games. Take your shot. Pick a winner.

Compare 2 models*
vs
Model
Effort
Sort by

Notes

*This isn't a rigorous experiment. Each prompt was run only once per model and effort, with no repeats, and the builds were made in different environments (local on macOS and cloud on iOS app). So read it as a fun side by side, not a benchmark.

**The Claude iOS app's Claude Code surface exposes only the model picker, not the separate /effort control in the in-app UI that the official docs describe for terminal and desktop. The iOS builds (Fable 5 and the base Opus 4.8) used the app's default, which was likely High.

***Fable 5.1 Ultracode and Opus 5.5 Max ran into session limits. Therefore, they were the only games that had intermediary prompts such as "Try again" and "I hit my usage limit while you were working, but it has reset now. Please continue from where you left off", sent automatically when session limits reset. The time indicated here is strictly the time the model spent working. Time spent idle while waiting for limits to be restored was excluded from the measurement.