Same prompt, different models

Orrery Bench

We give different AI models the exact same one-line prompt and publish what each one builds. Since the prompt never changes, the differences between these simulations come from the models themselves.

The prompt — word for word, every run
Same promptCopied exactly, typos and all. Nothing added.
One shotNo follow-up hints, fixes, or steering after the prompt.
Built aloneNo reference build and no code shared between runs.

At a glance

Some runs are community builds made by other people with the model named. Their cards give credit and explain the permission to host them here.

The runs

Side by side

How each build handles the physics and the engineering. Rows where every run agrees are hidden automatically.

For maintainers: adding a run

Put the build in its own directory, then add one object to ENTRIES at the top of this file. Everything on the page renders from that list, so nothing else needs editing.