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

# Credits & Billing

> How credits work, what bills (and doesn't), how to read the accounting headers, and where to manage quotas.

ByteSpike bills in **credits** where `1 USD = 1,000,000 credits`
(micro-USD precision). Per-token / per-call rates are quoted in dollars
in the [pricing table](https://bytespike.ai/pricing) and refreshed
nightly from the production gateway.

## What costs what

| Surface                                                                                                 | Pricing model                                                                                                      |
| ------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------ |
| Text endpoints (`/v1/messages`, `/v1/chat/completions`, `/v1/responses`, `/v1beta/.../generateContent`) | Per 1M input + per 1M output tokens. Cache reads discounted, cache writes at 1.0× (most models) or 1.25× (Claude). |
| Image endpoints (`/v1/images/generations`, `/v1/tasks/submit` for batched/async)                        | Per image. Discrete per-call cost.                                                                                 |
| Video endpoints (`/v1/tasks/submit` for Veo)                                                            | Per second of output. Failed renders are free.                                                                     |
| Utility (`/v1/models`, `/v1/balance`, `/v1/usage`, `/v1/tasks/{query,cancel}`)                          | Free.                                                                                                              |

The authoritative live rate is always [bytespike.ai/pricing](https://bytespike.ai/pricing).

## Failures don't bill

Any non-2xx response is free, regardless of which model failed or how
far into the request the failure occurred. This is a hard contract —
the `X-Quota-Remaining-Credits` header doesn't move on a non-2xx.

The narrow exception: if you cancel a video task **after** rendering
has started (status `running`), the partial seconds
rendered may bill depending on the model's own refund policy. The
`credits_used` field on the cancel response is authoritative. See
[`tasks/cancel`](/api-reference/utility/tasks-cancel).

## Accounting headers

Every response carries the quota envelope (success and failure both):

| Header                          | What it means                                                                                                   |
| ------------------------------- | --------------------------------------------------------------------------------------------------------------- |
| `X-RateLimit-Limit`             | USD cap of the rate-limit bucket closest to constraining you (the tightest of `rate_limit_5h` / `_1d` / `_7d`). |
| `X-RateLimit-Remaining`         | Budget left in that bucket.                                                                                     |
| `X-RateLimit-Reset`             | Unix timestamp when the bucket resets.                                                                          |
| `X-Quota-Remaining-Credits`     | Lifetime credits remaining on this key (USD). `0.00` = key's `quota` cap reached.                               |
| `X-Org-Quota-Remaining-Credits` | Org wallet remaining, on org-owned keys.                                                                        |

The actual cost of an individual request isn't in the response headers
— it's available via [`GET /api/v1/usage`](/api-reference/account/me-usage),
which returns one row per call with `prompt_tokens`, `completion_tokens`,
and the final billed `credits`.

## Pre-flight budgeting

For "this will cost \~\$X, confirm?" flows:

1. Compute a worst-case from `max_tokens` × output rate + prompt size × input rate using the [pricing table](https://bytespike.ai/pricing).
2. Compare against `X-Quota-Remaining-Credits` from a prior call (or hit [`/v1/balance`](/api-reference/utility/balance) — free).
3. If the estimate exceeds your budget, don't send the request.
4. After the request, reconcile actual cost via `/api/v1/usage`.

ByteSpike does not currently emit a pre-flight estimate header — the
quota headers are post-billing only.

## Quota cliffs

A key stops serving requests when **any** of the following is true:

* `X-Quota-Remaining-Credits = 0` (key's `quota` cap reached)
* `X-RateLimit-Remaining = 0` (key's tightest rate-limit bucket exhausted)
* Org wallet is empty (org-owned keys)

The gateway returns `402 insufficient_balance` (OpenAI envelope) or
`permission_error` (Anthropic envelope). To raise the cap, edit the
key in [Console → API keys](https://console.bytespike.ai/keys) or
top up the org wallet.

## Top-up and subscriptions

Top up the org wallet at [Console → Billing](https://console.bytespike.ai/billing).
The minimum top-up is \$5; bonuses scale up to +11.2% on larger packs.

Subscription tiers (Pro / Max / Enterprise) bundle credits with
higher concurrency and priority. See [bytespike.ai/pricing](https://bytespike.ai/pricing)
for the live tier table.