Skip to main content
fusion sends one prompt to two or three models at once, then a judge model synthesizes their answers into one. It is the command to reach for on a decision where a single model’s confident answer would hide that the question is genuinely open.
Fusion requires a paid plan. See What Paid Unlocks.

​
Syntax

You name two or three participant models as positional arguments, and they must be distinct. More than three is rejected with a clear message. Run fusion with no arguments at all and it walks you through picking them instead of erroring.

​
Options

​
How Privacy Levels Mix

Participants must share a compatible privacy level:
  • Anonymous and Private can mix in one Fusion.
  • Confidential must stay in a Confidential-only Fusion. Combining a Confidential model with a lower level is refused, so a synthesis never quietly downgrades the privacy of a Confidential participant.

​
Reading the Results

A Fusion result has two parts:
  1. Each participant, listed separately, with the model, timing, tokens, and whether web search was used.
  2. A synthesis: one answer combining their strengths. The judge resolves disagreements where it can, and flags them briefly where it cannot, when they are material to the answer.
Seeing each participant’s answer next to the synthesis is what tells you whether a decision was settled or contested.

​
Example

The result lists each model’s answer, then a synthesized recommendation with the tradeoff and a note on where the three models diverged.

​
Next

Fusion participants are always explicit, so picking them well matters. See Choosing a Model.