Consensus

Coming soon. Consensus through the API is not open yet.

In the app, Consensus asks 2 to 6 models the same question. Each answers on its own, then they read and critique each other's answers without knowing whose they are, and one model writes the result:

• Synthesized answer – one answer built from all of them.
• Model agreement – the points the models share.
• Areas of disagreement – where they differ, and who holds which position.
• Confidence based on model positions – how many models took the same position and how firmly, for the formats where that can be measured.

Formats: General, Research, Coding, Creative, Decision and Debate; the Idea validation template runs a Decision consensus on a product idea. The estimate is shown before you start, and the answers show once two models have answered. A session in which fewer than two models answer, or that fails on our side, is not charged. One you stop, or whose synthesis fails, after two or more models have answered is charged for the model calls already completed.

A consensus is a structured comparison of model answers, not advice, and it does not tell you what is true.

Until the API opens, you can build a simple version yourself: send the same question to several models with Chat, then ask one of them to compare the answers.