What each plan costs, what it actually includes, and what the next one up buys you. Pricing owns the plan terms — if a number here disagrees with it, that page is right and this one is a bug worth reporting.
To subscribe, go to Join. This page is the explanation, not the checkout.
| Plan | Monthly | Annual | Annual works out at |
|---|---|---|---|
| Free | $0 | — | — |
| Plus | $20 / mo | $200 / yr | $16.67 / mo saves $40 |
| Pro | $45 / mo | $450 / yr | $37.50 / mo saves $90 |
| Q‑Teams | $30 / seat / mo | $300 / seat / yr | $25 / seat / mo saves $60 |
Annual is billed once, up front. The “works out at” column is arithmetic, not a separate price you can pay — there is no $16.67 monthly plan. Every annual plan is two months free against paying monthly for a year.
Subscriptions renew automatically until cancelled. Cancelling takes effect at the end of the period you have already paid for. The full terms are in the Terms of Service, §04 and §06.
Q‑Teams requires at least three seats. That is a minimum number of seats, not a $90 floor: you are buying three or more seats at $30 each, and the price per seat does not change with volume. Stripe enforces the minimum on the subscription itself, so it holds on every later change too, not just at checkout.
Seats can be added or removed at any time; changes are prorated against the remainder of the period.
Not a number of questions. A deliberation runs several models, so its cost depends on how hard the question turns out to be and how much the panel had to argue about it. Plans are sized in roughly how much deliberation time is included per month — about 20 hours on Plus, about 50 on Pro.
“Roughly” is doing real work in that sentence. Ten easy questions cost a fraction of one genuinely contested one, so nobody — including us — can convert that allowance into a reliable question count. Every answer carries a receipt showing what it actually consumed.
Turbo credits top up your account without changing your plan: $10, $25 or $100, one-time. They do not expire, they are consumed as you use the Service, and they have no cash value. Auto‑reload is off unless you switch it on in Settings; while on, it charges the smallest pack when a request would otherwise be blocked, capped at three reloads in any 24 hours.
The honest version, because “better models” is not a specification.
Every plan has a ceiling on how capable an engine it can reach, and the ceilings are cumulative — Plus reaches everything Free does, Pro reaches everything Plus does.
Pro is not a different universe of models. It is a comparatively small set of the most capable ones, plus Einstein Mode, more included time, unlimited Studio, and more storage. If your questions are mostly ones Plus already handles well, Plus is the right plan — and we would rather you stayed on it than paid for headroom you do not use.
Which engines run on any given question also depends on the Express / Foundation / Frontier control next to the composer, which is a per-question choice rather than part of your plan. That is explained on Help.
The API is billed per deliberation against an organisation
balance rather than a monthly plan. POST /v1/estimate prices a
question before you commit to it and is always free.
A call is billed as tokens plus a per-deliberation surcharge. The surcharge is the panel itself — several models answering, critiquing and being judged — and it is the larger part of a typical bill.
| Direction | Rate |
|---|---|
| Input | $2.00 per 1M tokens |
| Output | $8.00 per 1M tokens |
| Model | Light | Medium | Deep | Ceiling per call |
|---|---|---|---|---|
quorum-standard |
$0.05 | $0.10 | $0.15 | $0.40 |
quorum-max |
$0.32 flat | $0.50 | ||
Light, medium and deep are decided by the classifier, not by you
— easy questions are charged less because they cost us less to run. The
classification is the same one /v1/estimate returns for free, so
you can see which band a question falls into before spending anything.
Production distribution as of 2026-08-20, across 3,947 classified questions:
46% light, 33% medium, 21% deep.
The ceiling is a hard stop on what a single call can bill. If a
deliberation would exceed it, you still get the answer and the response comes
back with finish_reason: "cost_cap".
A medium-difficulty question on quorum-standard with a 400-token
prompt and a 700-token answer: $0.0008 input + $0.0056 output + $0.10
surcharge = about $0.107. The tokens are rounding error; the
surcharge is the price.
Modes carry their own pricing, so a Mode you build or buy may differ from both
rows above. GET /v1/models returns the exact figures your key is
entitled to, including any negotiated override, and is authoritative over this
page for your own organisation.