Skip to content
Pay as you go

Pay for the tokens you actually use.One prepaid balance. Published per-model rates.

Add AI credit when you need it, call the models your key allows, and see every request — its tokens, its rate and its charge — in the dashboard. This lane has no recurring charge and no monthly commitment.

Already have an account? Sign in and add credit from the dashboard.

How a top-up becomes usage

  1. 01Top upYou name the amount, above the minimum shown at checkout.
  2. 02AI credit is addedThe bonus tier for that amount is applied by the server and confirmed at checkout.
  3. 03Requests are meteredEach call is priced with the model's own input, output and cache rates.
  4. 04The balance is debitedThe settled charge is what the request cost — shown per request.

Unused prepaid balance does not expire, and no charge is made on a schedule.

Mechanics

How the prepaid balance works.Top up once, spend at the model's rate.

The balance is money, not a package of pre-counted requests. Its value is spent by the token, at whatever rates the models you call are configured with.

Choose an amount

A top-up is free-amount: you choose the amount — one of the quick amounts or a value you type — as long as it meets the minimum shown at checkout. There is no fixed package to pick.

The bonus is applied server-side

Larger top-ups earn a configured bonus tier. The tier that applies to your amount — and the final AI credit it produces — is calculated and shown at checkout before you pay.

Spend it at the published rate

Every request is metered with the selected model's configured token rates and settled against the balance. The charge is the cost of the request that actually ran.

Live top-up configuration

Bonus tiers, read from the platform.

Every minimum and percentage below is served by the same configuration the checkout applies. Tiers are cumulative: the largest threshold your amount meets is the one used, and checkout shows the exact AI credit before you pay.

Loading live top-up tiers…

Rate card

What a request is billed for.

Requests are priced in token categories against the selected model's configured rates, all quoted per 1 million tokens. The catalogue is where the per-model numbers live.

Input tokens

Prompt tokens are charged at the model's published input rate per 1 million tokens. The rate comes from the model's current pricing row — the same row the billing engine reads.

Output tokens

Completion tokens are charged at the model's published output rate per 1 million tokens. Input and output are priced separately, so a long prompt and a long completion never share one blended number.

Cached input

Cache reads and cache writes are their own priced categories, metered against the model's configured cache-read and cache-write rates. Where a rate card does not price a cache category, it falls back to that model's input rate — cached tokens are never assumed to be free.

Public model catalogue

The live model list with input and output rates per 1M tokens, cached-token rates where the rate card publishes them, and context windows.

Open /models

In your dashboard

Signed in? The dashboard catalogue shows the same live rows inside your account, including your price and any original price the server publishes.

/dashboard/models
Your charges

Every request is on the record.

The usage page lists each request with the model it used, its prompt and completion tokens, cache read and cache write tokens, the credits charged, the status and the latency. Filter by model and date range, then export the range as CSV.

  • Per-request token counts, including the cache categories
  • Credits charged for each request, and usage by model
  • Status and latency, so a failure is visible next to a success
  • CSV export of the filtered range
/dashboard/usage
Modelthe model the request used
Tokensinput · output · cache read · cache write
Creditswhat the request was charged
Statusthe response status, with latency

Column structure shown; actual rows and amounts appear only in your dashboard.

Guardrail

When the balance runs out

The balance is checked before a request is sent to a model. If it cannot cover the request, the API refuses it with HTTP 402 and the error code insufficient_credits. The request never reaches the model, so there is nothing to pay for it.

Top up and retry the same call. The failed request is never charged, so a retry only costs what the retry itself uses.

POST /v1/chat/completions402
{
  "error": {
    "message": "Balance empty — buy more credits in the dashboard",
    "type": "insufficient_credits",
    "code": "insufficient_credits"
  }
}

Response shape shown for the OpenAI-compatible endpoint; the error code is the same on the Anthropic-compatible endpoint.

Catalogue

Current models. Current rates.One source, two views.

Prices are never a static table on this site. The public catalogue and the dashboard catalogue both render the live model rows the billing engine prices against, so a rate change is visible in both places at once.

Public catalogue

Anyone can check which models are published, what each one costs per 1M input and output tokens, and which cached-token rates exist — before creating an account.

Browse models and rates

Dashboard catalogue

Inside your account the catalogue is the same live data, next to your keys and your balance — so you can compare a rate before you spend on it.

Open dashboard models

Start with a top-up.Spend it at published rates.

Create an account, add AI credit, and send your first request against the model catalogue. Your balance and every charge stay visible in the dashboard.

Prepaid balance · no recurring charge · exact price shown at checkout