Ask once.
Hear from three.
Get the verdict.
Put the same question to up to three models at once. When they finish, the council tells you where they agree, where they differ, and which answer to trust, so you read one synthesis instead of three essays.
Three answers.
One reading.
Every step is an ordinary request on your balance, priced to the fraction of a cent like everything else in the chat.
Pick up to three models
Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, or let auto choose. Same prompt, same moment, one balance.
They answer in parallel
Each model streams its own column. One being slow, or wrong, does not hold up the others.
The council reads all of them
One more model reads every answer and writes three things: where they agree, where they differ, and which one to trust for this question.
Confidence
is not evidence.
A single model answers with the same certainty whether it is right or not. Three of them, asked independently, cannot all be confidently wrong in the same direction very often, and when they split, the split is the information.
Decisions with a wrong answer
Architecture, data modelling, a migration plan. Three independent takes surface the assumption one of them is quietly making.
Code you are about to ship
Paste the function. Where all three flag the same line, it is the line. Where only one does, the verdict says whether to listen.
Anything you would otherwise ask twice
If your habit is to ask a second model when the first sounds too sure, this is that habit, done for you, with a summary at the end.
Convene your
first council.
Free starter credits on signup cover it. Turn on Compare, add a second model, and the council switch is right there.