One question.
Every model weighs in
before you get an answer.

Every prompt runs across multiple models in parallel. A judge model reconciles the disagreements and hands you the answer worth trusting — not just the first one that came back.