Pricing
How you pay for the gateway — subscription plans, x402 wallet payments, and how usage is metered per capability.
Every call against the gateway is metered — the unit differs per capability, but the accounting
is always usage-based underneath. How you settle that usage is a separate choice from how it's
measured, and the gateway supports two live payment paths side by side on the same vk_ key.
Two ways to pay
Subscription plans
A recurring plan grants a weekly cost budget plus per-unit allowances. Purchased and managed in the Console.
x402 wallet payments
Pay per request straight from a wallet — sign a USDC maximum on Base, the gateway settles only what you use.
A workspace on a subscription plan authenticates with its vk_ key or a short-lived plan
JWT; a wallet paying over x402 needs neither — the signed
payment on Base is the auth. The two paths
aren't exclusive: a subscribed workspace that opts in to extra usage can spill past its plan
window onto a prepaid balance, and a workspace with no plan at all can run purely on prepaid
balance, topped up ahead of time instead of signed per request.
Prepaid balance is the pay-as-you-go fallback
Balance is what every request lands on when there's no plan window to draw from — a workspace
with no active subscription, a plan request that overflows its weekly window (only if extra
usage is enabled), or an x402 payment settling in batch-settlement mode. It's deducted per
unit at the same rates the model catalog will publish.
What's metered, and in what unit
Every capability bills a different unit — there's no flat per-request or per-second-of-wall-clock charge:
| Capability | Endpoint(s) | Billed unit |
|---|---|---|
| Chat / text generation | /v1/chat/completions, /v1/responses | Tokens (prompt + completion) |
| Embeddings | /v1/embeddings | Tokens (prompt only — no completion) |
| Audio speech (TTS) | /v1/audio/speech | Input characters |
| Audio transcription/translation (STT) | /v1/audio/transcriptions, /v1/audio/translations | Audio seconds uploaded |
| Images | /v1/images/generations, /v1/images/edits | Per generated image (n: 4 bills as four) |
| Video | /v1/videos | Generated seconds, scaled by the job's fps |
Token-metered capabilities (chat, embeddings) report the exact count in the response usage
block, so you can reconcile after every call. The count/duration-metered ones (audio, images,
video) reserve an estimate up front and reconcile to the real value the model reports once the
job finishes — a job that fails or produces nothing settles for nothing.
No published per-unit rates yet
This page describes how usage is measured, not the dollar rate per token, character, second, or image — those live per model and are on the way at Models. For a subscription, that rate mostly stays invisible: the plan window is denominated in cost, not a raw count, so a heavier model just drains the window faster. For prepaid and x402, the per-unit rate is what actually gets debited or signed.
Next steps
- Subscription plans — the recurring, cost-denominated weekly budget and how overflow works.
- x402 — pay per request from a wallet, no plan or
vk_key required. - Quickstart — get a
vk_key and make your first call.