Referee.chat

Referee.chat sets up debates between multiple AI models on a question or claim, with a referee — configured by the user — judging the exchange on evidence rather than which model sounds most confident. A user states a goal, up to six AI models take debating seats, and the panel searches academic sources and cross-checks claims as part of the process; the referee decides when the debate is finished and delivers a verdict. The site offers several referee standards to choose from — described as careful-expert, publication-standard, formal-proof, and decision-grade — letting a user pick how rigorous the judging should be depending on whether they want a quick sanity check or something closer to a reviewed conclusion. It's aimed at people who want to stress-test a claim or compare how different models reason about the same question, rather than get a single model's unchallenged answer. A related product, theorem.chat, applies the same debate-and-referee approach specifically to mathematical proofs formalized in the Lean proof assistant. One limitation: the site's own homepage does not disclose pricing details or which underlying AI models power the debates, so cost and model transparency have to be checked on its separate pricing and models pages before relying on it for serious use.