Every answer, checked before you see it.

Your question goes to several models at once. The answers get scored, compared for agreement, and reviewed by a referee. You get the one that survives.