How billing works

One prepaid balance for Claude, GPT, Gemini and Kimi

Billing on apiToken.sale is prepaid and pay-as-you-go: you fund a balance in advance, and API requests draw it down — there is no monthly invoice and no customer subscription. The same balance covers every supported Claude, GPT, Gemini and Kimi model, so you never manage four separate provider accounts or four separate bills.

You top up any whole-dollar amount. The balance never expires, which changes how you should fund it: there is no reason to preload a large sum "just in case", and no reason to hoard either. Add what the next few weeks of work need, and top up again when the dashboard says you are running low. Idle time costs nothing — a balance you do not spend this month is still there next quarter.

See also: How Claude API pricing works

How a single request becomes a charge

Every call is priced from the usage the provider actually reports, not from a flat per-request fee. The router converts the response's usage fields into official provider spend, leg by leg:

• Input tokens — your prompt, system instructions and conversation history, at the model's official input rate.
• Output tokens — the model's reply, at its official output rate, which is usually higher than input.
• Cache legs — cache writes and cache reads are metered separately from regular input, and cache reads are much cheaper.
• Model-specific buckets — long-context surcharges or image legs where a model's rate card defines them.

Because metering follows each provider's own rate card, a Claude request is priced exactly like Anthropic would price it and a Gemini request exactly like Google would — before your discount enters the picture. Streaming responses are metered the same way as non-streaming ones; the final usage arrives with the stream, and the charge is identical.

Where the 50% discount enters the math

The discount is applied after official spend is computed, never before. Official spend comes first, then your flat 50% B2C discount is subtracted across every supported provider, and the net amount is deducted from your prepaid balance. A worked example on Claude Haiku 4.5 (official $1 per million input tokens, $5 per million output tokens):

The same order of operations applies to GPT, Gemini and Kimi requests against their own official rate cards. Prompt caching compounds with the discount rather than replacing it: a cache read is already cheaper at the official rate, and the 50% comes off that cheaper figure too.

Funding the balance: cards, crypto and amounts

At checkout you can pay by bank card or with cryptocurrency through a secure payment provider; crypto top-ups credit once the network confirms the transaction. Either way the funds land as prepaid balance in your account and are spent only when requests run — you can switch between card and crypto per top-up.

If the bonus or a top-up does not behave as expected, support is available in English and Russian via Telegram, or by email at apitokensale@gmail.com. Refund handling is processed through the original payment provider.

What happens when the balance hits zero

Requests start failing with HTTP 402 billing_error and the message insufficient balance or key spending limit reached for this request — your key is not revoked and your account is not closed, the calls simply stop going through until you top up. Any whole-dollar amount restores service immediately, so the fix is a checkout visit, not a support ticket. Do not retry 402 as if it were a 429.

curl https://router.apitoken.sale/v1/messages \
  -H "x-api-key: sk-pool-•••" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-4-5",
    "max_tokens": 64,
    "messages": [{"role":"user","content":"ping"}]
  }'

# With an empty balance the response is:
# HTTP 402 billing_error — insufficient balance or key spending limit reached for this request

For unattended workloads — agents, cron jobs, CI evals — check the dashboard balance before long runs. A 402 mid-queue is a failed job, and retries cannot fix an empty balance the way they fix a rate limit.

Auditing where the money went

Dashboard Usage shows settled spend by model, provider, day, and customer key. The ledger remains the money journal of charges and top-ups. For a request-level audit, call GET /activity with a customer key; it returns model, provider, tokens, charged amount, public error, and optional agent session grouping.

To plan a top-up rather than reconcile one, estimate the workload first: per-model official and discounted rates are listed on the model pages, and the cost calculator turns a prompt-plus-reply token estimate into a dollar figure before you spend anything.

Per-model rates, cache pricing and context windows →

Estimate a workload with the Claude API cost calculator →

Frequently asked questions

Related guides