One prompt.
Every model, live.

Race the same prompt across two to four models. Tokens stream side by side, and every answer ends with its time to first token, total time, cost and a signed receipt.