One prompt. Every model.
One try each.
FeelBench gives the same prompt to a lineup of AI models and publishes whatever each one builds, exactly as written, no retries. Compare the results side by side, or guess which model made what.
Just for fun. Not a leaderboard.
Or take one model at a time
Every toy a single model built, in one place — then flip it against any other model, toy by toy.
One frozen prompt
Every model gets the exact same words — read them on any toy's page and run the dare on your own model if you like.
One try, no edits
Fresh session, no follow-ups, no fixes. What the model wrote in one pass is what you're playing — including the broken bits.
You're the judge
No scores, no rankings, no AI graders. You play the toys and draw your own conclusions. The shelf keeps growing whenever we think of a new dare.