Tokens and billing

Claud is prepaid. You hold a balance of Claud tokens, every request debits exactly what it used, and you can always see the cost of a request in its response. There are no invoices after the fact and no way to run up a bill you did not fund.

Claud tokens#

A Claud token is a fixed unit of account: 1 Claud token = $0.000002, so 500,000 Claud tokens = $1.00. The rate is returned by GET /v1/billing/balance as token_rate.

Claud tokens are not the same as the model tokens counted in usage.prompt_tokens and usage.completion_tokens. Model tokens measure text; Claud tokens measure money. The conversion goes through each model's published price:

Text
cost_usd   = prompt_tokens        × input_price  / 1,000,000
           + cached_prompt_tokens × cached_price / 1,000,000   (instead of input_price for those tokens)
           + completion_tokens    × output_price / 1,000,000   (reasoning tokens are completion tokens)

claud_tokens_debited = ceil(cost_usd / 0.000002)

Example. A claud-5.1 request with 1,200 prompt tokens (400 of them cached) and 300 completion tokens at $0.75 / $3.00 / $0.075 per million:

Text
input   800 × 0.75  / 1e6 = $0.000600
cached  400 × 0.075 / 1e6 = $0.000030
output  300 × 3.00  / 1e6 = $0.000900
total                      = $0.001530  →  765 Claud tokens

Every response reports both numbers: usage.claud_tokens_debited and usage.claud_cost_usd (non-streaming) or the final fable event (streaming). Current prices per model are on the Models page and in GET /v1/models.

Prompt caching is automatic

When a prompt prefix repeats (a long system prompt, a document, earlier turns of a conversation), the provider serves it from cache and Claud bills those tokens at the much lower cached rate. usage.prompt_tokens_details.cached_tokens shows how many tokens were cached. Keep stable content at the start of the prompt to maximise hits.

How a request is charged#

  1. Estimate. Before calling the model, Claud estimates the prompt size and assumes the maximum possible output (max_tokens or the model default) to compute a worst-case cost, adds a 20% safety margin, and reserves that many Claud tokens from your balance. If you do not have enough, the request fails immediately with 402 insufficient_balance and nothing is charged. Reserved tokens appear as reserved_tokens in your balance while the request runs.
  2. Run. The model generates the response.
  3. Settle. The reservation is replaced by the exact cost based on the provider's real token counts. The difference is released within milliseconds.

Consequences worth knowing:

  • Setting a realistic max_tokens lets you make requests with a smaller balance, because the reservation is smaller.
  • Requests that fail before producing output (4xx, provider errors, rate limits) are never billed.
  • If a stream is cancelled or fails part-way, you pay only for tokens actually generated.
  • A small minimum balance (1,000 Claud tokens, $0.002) is required to start any request.

Where tokens come from#

Your balance has two buckets, drawn down in this order:

BucketSourceExpiry
SubscriptionMonthly allowance granted by a paid plan (12M on Starter, 39M on Pro, 140M on Power, 525M on Business; none on pay as you go)Unused allowance expires at the end of each billing period and when you change plans.
PurchasedOne-off token packages you buyPackages may carry an expiry (shown at purchase); most do not.

GET /v1/billing/balance and GET /v1/account break the balance down by bucket, and GET /v1/billing/ledger returns every credit and debit as a double-entry ledger you can reconcile against your own records.

Plans#

Plans set your monthly allowance, rate limits, maximum context and which models you can use. You can compare them on the Pricing page. Upgrading takes effect immediately: the new allowance is granted at once and the higher limits apply to the next request.

Token packages#

Packages are the way to add headroom beyond your plan or to use Claud without a subscription. Larger packages include bonus tokens. Buy them under Billing.

Paying#

Claud accepts cryptocurrency payments directly, with no third-party processor and no KYC: there is no identity verification, no card on file and no billing address to enter. An email address is the only thing an account needs.

  • Networks and assets: Bitcoin (BTC); Ethereum, Base, Arbitrum, Polygon and BNB Smart Chain (native coins plus USDT and USDC where available); Tron (TRX and USDT).
  • How it works: choose a plan or package, pick an asset, and you receive a unique deposit address and exact amount quoted at the current market rate. The quote is valid for 30 minutes. Send the payment; tokens are credited automatically once the network reaches the required confirmations (from 1–2 blocks on Bitcoin to about a minute on fast EVM chains).
  • Underpayments within a small tolerance (0.5%) are accepted as paid. Larger shortfalls are held for review; open a support ticket with the invoice ID and we will credit the amount received. Overpayments are credited in full at the quoted rate.
  • Every payment and the tokens it produced are listed under Billing and in the ledger.

Claud's payment addresses are derived from watch-only keys: the platform never holds private keys, and funds go straight to cold storage.

Low balance and cost protection#

  • You receive an email when your balance drops below 100,000 tokens ($0.20), and the dashboard shows a warning. Adjust the threshold or turn it off in Settings.
  • Because the system is prepaid, there is no overdraft: when the balance runs out, requests fail with 402 insufficient_balance until you top up. Nothing else changes; keys and conversations remain intact.
  • Watch x-claud-balance on non-streaming responses (or balance_tokens in the stream's fable event) to top up before your users notice.

No refunds#

All purchases are paid in cryptocurrency and are final: there are no refunds, cancellations or chargebacks once a payment confirms on-chain. Double-check the network and amount before you send. If you believe a request was billed incorrectly, send the request_id to support; every request's cost breakdown is recorded and we will correct any error.

Reference#