These are the minigames from the paper. They keep moving while you think, just as they do for the models.
The strongest model, GPT-6 Astra, solves only 9% of them in real time.
9%Best model, live
45%Author, first try
0Games you solved
These minigames need a mouse and keyboard. They work best on a computer.
No minigame matches these filters.
Time limit
You solved
Difficulty
Controls
Each play gives you a new challenge from the benchmark's generator. The benchmark's own grader checks your
answer in the browser, so the first check takes a few seconds while it loads.