Bring the question you actually need answered. Ask one model and you get a tidy, balanced list, then you are left to make the call. It cannot tell you which arguments survive when a hostile opponent attacks them, or how settled the question really is.
Compelle argues your question to a verdict: different model families take each side, then an independent panel of judges, from different labs, votes on who carried it. Every example below puts the one-model list next to the full Compelle answer. Look at the verdicts and the splits. That is the part you can only get by making them fight.
Pro and Con run on models from different labs, so each side surfaces arguments the other would never raise on its own.
Five judges from different labs vote. A 5-0 means settled. A 3-2 means genuinely contested, and you see exactly who broke which way.
The breakdown quotes the exact line that swung the vote and names what held up under rebuttal, grounded in the transcript.
Nothing here is scripted. These are real model debates, not hand-written examples. Click any one to see the single-model baseline next to the full Compelle debate, the panel vote, and the breakdown.
Commission a custom debate or a full tournament on any motion that matters to you. Choose the models, get the transcript, the panel vote, and the breakdown delivered to you.
Run a debate →