A Mode is a saved configuration of how Quorum deliberates. Not what it knows and not what it is told — who sits in the seats, how hard they argue, what it costs, and how it presents itself.
When you call the API, the model string does not name an engine. It names a Mode.
Every Mode is the same four blocks. Nothing else.
| Block | Decides | Example of a real setting |
|---|---|---|
engine | Who sits in the seats | Seat 1 Anthropic, seat 2 Google, seat 3 OpenAI |
deliberation | How hard they argue | Whether a synthesis pass always runs, or only when the question earns it |
commercial | What it costs to call | A per-call ceiling, and a surcharge that varies with difficulty |
presentation | How it appears | Its label, and whether engines are named to the reader |
The most important setting in a Mode is the one most people never look at: how seats are filled.
Quorum’s default Mode assigns seats by family first, then picks the strongest available engine within each family for the question at hand. The consequence is the entire product thesis, made mechanical: you always get three differently-trained models, never three flavours of the same one.
A Mode that filled seats purely by capability score would, on a good day for one lab, quietly hand you the same lineage three times and call it a panel. Family-first assignment is what stops that.
Some questions genuinely do not need breadth, and a Mode built for those can say so. What matters is that the choice is recorded in the Mode and visible on the listing, rather than being an invisible default nobody can inspect.
Deliberation is not all-or-nothing. A Mode sets a policy:
A Mode also caps how many rounds of argument are allowed. That is a ceiling, not a target: the loop stops early as soon as a round stops producing genuine movement, so most deliberations finish well below the cap.
Two commercial settings matter to anyone calling a Mode.
A Mode may also carry a flat fallback surcharge, used when difficulty cannot be determined. If you ever see a Mode charging its flat rate on questions that are obviously easy, that fallback is doing the work and something upstream is wrong. Full plan and call pricing lives on Pricing, explained.
Worth stating, because the word invites the wrong assumption.
Some things about a Mode are decided by the server rather than its maker, and the reasons are the same in every case: a number you can set yourself is a number nobody else can trust.
A Mode is a claim about how a question should be argued. The parts a maker controls are the interesting ones. The parts they do not control are what make the claim worth anything.
Browse what exists on the Modes. Build one at Build a Mode — a signed-in account is the whole requirement. If you are weighing whether to publish, Creators covers what that involves. For calling a Mode over the API, see the reference; for which of them your key may reach, GET /v1/models.