> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trylimitless.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Credits and billing

> Understand prepaid balances, request reservations, and settlement.

Limitless uses prepaid USD credits for inference. In **Billing**, inspect **Available credits**, **Reserved credits**, **Total credit balance**, and **Credit history**. The **Model prices** table shows the configured rates.

<Note>
  The current site uses **Stripe test mode** and sandbox credits. Test checkout does not collect real money; inference still reaches real models and deducts sandbox credits. Check [Billing](https://trylimitless.ai/app/billing) for the configured model prices. Sandbox balances and test payments do not become paid production credit.
</Note>

## Add credits

Owners and admins can select **Add credits** when Stripe checkout is configured. Choose an amount, continue to Stripe, complete payment on the hosted checkout page, and return to Billing. **Add test credits** identifies Stripe test mode; a test checkout does not collect real money.

Credits are added only after the backend verifies a successful payment event. Returning to the console, a success query parameter, or a screenshot of a payment does not itself add credits. The console shows **Checking payment** until it finds the confirmed credit, then **Credits received**. Select **Refresh balance** before trying another payment if confirmation is delayed.

## How a request affects your balance

1. **Reserve:** Before dispatch, the gateway checks available funds and reserves a conservative amount for the model's configured maximum input window at the uncached input rate, plus your bounded output limit multiplied by Chat's requested completion count (`n`, default 1).
2. **Run:** The request uses a concurrency slot while the backend generates its response.
3. **Settle:** With complete provider usage, the final charge uses the input and output rates captured for that request. Unused reserved funds become available again.

A reservation is a temporary hold, not the final token charge. Because the input reservation uses the model's configured maximum input allowance, the required available balance can be larger than the expected cost of a short prompt. Reducing the output limit reduces only the output portion of the reservation.

## Token pricing

Input and output tokens can have different USD rates. A model can also have a cached-input rate. When the provider reports cached tokens, those tokens use the configured cached rate. The conceptual charge is:

```text theme={null}
(input tokens − cached input tokens) × input price per million / 1,000,000
+ cached input tokens × cached input price per million / 1,000,000
+ output tokens × output price per million / 1,000,000
```

The backend accounts in integer microdollars, rounding the combined input, cached-input, and output charge up once to the next whole microdollar. One USD is 1,000,000 microdollars. Your request's recorded price is used for settlement, so a later price change does not rewrite that request's rates.

There is no promise of a cache hit or a fixed cache discount. If the provider does not report a cached count, the system cannot infer it from repeated text.

## Failures and unknown usage

A failed request can still consume tokens. If the provider supplies usage, those tokens can be charged. A known rejection before generation can release its reservation.

When you use Stop or disconnect before the request finishes, confirmed final token usage is charged normally. If the gateway observes your cancellation without receiving final usage, it releases the reservation and makes no charge for that request. Activity keeps the token counts unknown; Limitless absorbs any unmeasured supplier cost. This does not mean the supplier's generation was free.

Unknown usage from timeouts, provider failures, platform shutdown or an interrupted accounting write can still leave a reservation held for reconciliation. Activity can therefore show unknown usage while Billing shows reserved funds. Those holds are not released merely because a request is old or its concurrency slot expires. Contact your beta operator with the request ID if a hold remains unresolved.

## Insufficient funds

Requests that cannot reserve enough available credit return `402` before inference. Add credits, reduce your output bound, or wait for existing reservations to settle. A monthly spending cap can also block admission even when the wallet has sufficient funds; see [limits](/limits).

## Programmatic balances and hosted tools

Customer keys with `credits:read` can call `GET /v1/credits`. Admin keys with `billing:read` can read `/v1/organization/balance`, `/v1/organization/usage/completions`, and `/v1/organization/costs`. Balances and reports belong to your Limitless organization.

Messages cache-read tokens use the cached-input rate. Cache-write tokens are counted with normal input. Hosted search and fetch currently have no separate Limitless fee; provider-reported model tokens are charged using the same pricing rules. Token-counting requests do not generate billable inference usage.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.