A check is the unit you pay for: one prompt, put to one engine, once — from 11 credits on the cheapest engine to 188 on the most expensive. Every run is costed before it starts.
1 prompt × 1 engine × 1 ask
Credits are the billing unit, and a check costs many of them — weighted by how expensive that engine is to query, so a cheap model costs less than a frontier one. The console prices the whole run before you commit to it.
Auto top-up stays off until you switch it on, with a ceiling you set.
80,000 credits pooled across every client workspace, up to 250 products, with the portfolio view and white-label reports. Bill your clients, not your credits. See how agencies run it →
Custom credit volume, unlimited products, SSO and roles, long retention, SLA and DPA — plus a quarterly review of what is actually moving your numbers.
Prices exclude tax. Paid yearly, every plan is ten months' price for twelve months' service, and you can switch cadence yourself in the billing portal. Out of credits early? Top up 5,000 for $39 or 25,000 for $149 — pack credits never expire — or turn on auto top-up with a monthly ceiling you set.
A grounded engine pays a per-search fee on top of the answer, and the priciest frontier model costs more than thirteen times the cheapest one. So rather than publish an averaged class price you would never actually be charged, here is the real per-engine cost. You are billed what the engine costs, never less.
No engine in this table is locked to a plan. All seven answer surfaces run on every plan, Free included, and what a plan buys you is credits and tracked products — never access to an engine.
| Engine | What it does | Credits per check |
|---|---|---|
| Grok | Frontier model, answers from training alone | 11 |
| Perplexity Sonar | Searches the live web while answering | 14 |
| Google AI Overviews / AI Mode | The answer surface itself, read through a SERP provider | 16 |
| Microsoft Copilot | Bing’s answer surface — the one engine here not built on Google’s index or a lab’s own crawler | 16 |
| ChatGPT · default | Searches the live web while answering — the tier a free ChatGPT user is served | 17 |
| Gemini | Searches the live web while answering | 30 |
| Claude | Searches the live web while answering | 91 |
| ChatGPT, paid browsing tier | The exact answer a ChatGPT Plus user sees. Agrees with the free tier on 94% of answers | 99 |
| Claude Opus | The most expensive model in the catalogue | 172 |
Prices move when the providers move theirs. The console always prices your specific run, on your chosen engines, before you commit to it — that quote is the number that binds, not this table.
A credit is the billing unit; a check is the work. One check is one question, asked of one engine, once — and it costs many credits, from 11 on the cheapest engine to 188 on the dearest. An engine that searches the live web to answer costs us a per-search fee on top of the answer itself, and the priciest grounded models cost several times the cheapest.
So we don't quote a single number here that would be wrong for your setup. The console prices every run before you start it: pick your prompts and engines and it shows the exact credit cost, against the balance you hold.
Agent and flow runs draw on the same credits, metered per tool call rather than per question. Because an agent decides for itself how many tools a job needs, its estimate is an honest range — one call if the first answer is enough, up to the run's ceiling — and both ends are shown before you start. A flow is priced as a whole graph, in the header beside the Run button. A run that calls no tools costs nothing and says so.
1,500 credits funds a genuine first baseline. Decide about a plan after you have seen the number.