Skip to main content
Your workspace pays for model usage from a prepaid balance. Requests use the published model rates, with no usage markup. Buying credits adds a 5% platform fee, plus tax where applicable. 1 credit = $0.000001. Requests served by your own provider key cost zero gateway credits; that provider bills you directly.

​
Topping up

In the console, select the workspace and choose add credits. The minimum purchase is $5. Credits become available after payment succeeds.
Create a hosted invoice, then open the returned invoice_url to pay:
The invoice API rounds fractional dollar amounts down to a whole dollar. Creating an invoice does not add credits; paying it does. Read GET /v1/billing for the balance, minimum purchase, and fee rate.

​
Avoid running out

Open auto top-up in the console, save a card, and choose a balance threshold and purchase amount. The amount must be at least the threshold and meet the minimum purchase. When your balance falls below the threshold, the gateway charges the saved card for the purchase plus fee and tax. Top-ups run asynchronously. Leave enough balance to cover traffic while payment completes. Failed charges can turn auto top-up off; check the message in the console, fix the card, and re-enable it.
A 402 can mean your balance cannot cover the estimated input cost, even when it is above zero. Output cost is settled after generation, and concurrent requests can overspend the balance. Prepaid credits are not a strict per-request or monthly spending cap. Bound concurrency and output length in your application.

​
Check what you spent

Open usage for totals or logs for an individual request. API callers can use:
/v1/account returns balance.credits and balance.usd. /v1/usage returns recent usage entries with a cost in both units. Request logs include failed requests and explain which provider served a BYOK or paid request.
For ordinary text without caching or special rates:
For example, at 0.15inputand0.15 input and 0.60 output per million tokens, 1,000 input and 200 output tokens cost 270 credits ($0.000270). This is an illustration; use the live model’s rates for an estimate.Actual billing accounts for each token class and rounds up once. GET /v1/models publishes cache-read and cache-write rates, one-hour cache writes, audio-input rates, and long-context tiers where they apply. A long-context tier can change the rate for the whole request.Other tasks have different units:
  • Transcription bills on audio duration.
  • Speech bills on characters or tokens, depending on the model.
  • Images bill on token usage, with separate image and text output rates where published.
  • Reranking bills on evaluated input tokens.
  • Evaluation bills on input tokens only; output tokens are free.
  • Provider-executed search can add a tool-call charge.
Read the unit alongside the price. An audio minute, a million characters, and a million tokens are not interchangeable.
Agent registration returns the starting balance in its response. Claiming an agent-created account may add a one-time verification bonus; eligibility details are in the agent quickstart.Some models require a completed credit purchase and return 403 billing_required without one. Saving a card or having promotional credits does not satisfy that requirement. Requests whose resolved routes are all BYOK are exempt.
If your workspace is configured for postpaid billing, usage draws against its credit limit and is invoiced. The console shows the billing cycle instead of prepaid top-up controls. A negative balance can be normal; requests are rejected when remaining credit headroom is insufficient. Auto top-up does not run for postpaid workspaces.