I mean, considering how many tokens their example prompt consumed, I wouldn't be surprised if it costs ~$2k/month/user to run