Claude API Top-Up Guide: Credits, Billing, and Pay-As-You-Go in 2026
If you are wiring Claude Code, Codex, or your own app to Claude models through a gateway, the first practical question is rarely the prompt — it is how to top up, what a call actually costs, and when the balance runs out.
This guide covers the full Claude API credit flow on CodeGateway: top-up options, how token billing works, the 1.05×–1.25× markup tiers, and how to verify your key after you pay.
How Claude API billing works (no monthly fee)
CodeGateway is pay-as-you-go. There is no monthly subscription and no minimum spend. Each API call deducts the actual token cost (tracked at micro-dollar precision). Balance never expires — unused credit stays on the workspace until you spend it.
Pricing has two layers:
Official base rate — the published per-token price for the model you call (Claude, GPT, Gemini, etc.).
Service markup — a multiplier on that base rate that drops as your rolling spend grows.
Tier (90-day spend) | Markup |
|---|---|
$0+ | 1.25× |
$20+ | 1.20× |
$50+ | 1.15× |
$200+ | 1.10× |
$500+ | 1.05× |
Higher volume → lower multiplier automatically. You do not need to apply for a discount.
Top-up options
You can fund a workspace balance with:
Card (Stripe) — standard cards; the processing fee (3.9% + $0.30) is shown as a separate line item, not hidden in the credit amount.
Alipay — available for users who prefer RMB checkout.
Bank transfer (CNY) — for larger amounts; invoices can be issued from the dashboard.
Common top-up amounts start from $10 (tiers such as $10 / $20 / $50 / $100 are exposed in the dashboard). Exact checkout amounts and any regional payment methods can change — the billing page is the source of truth.
$2 trial credit: new users who bind a card receive $2 free credit so you can verify the full call path without a separate invoice. There is no automatic signup credit without a payment method.
What one call costs (worked example)
A $20 top-up at the starter 1.25× markup buys roughly $20 ÷ 1.25 = $16 of upstream token cost. If the model’s base rate is $3 per million input tokens, that is on the order of 5+ million input tokens before the credit is gone (output tokens are priced separately and usually higher).
Your dashboard usage log shows per-request cost after the call settles, so you can reconcile the math against real traffic.
How to top up (step by step)
Sign in and open Dashboard → Billing (workspace wallet).
Choose an amount and payment method (card / Alipay / bank transfer).
Complete checkout — the balance appears on the workspace wallet.
Create or reuse an API key (
sk-cg-…).Point your client at
https://api.codegateway.dev(do not append/v1yourself for Claude clients) and send a small test request.
Verify the key after top-up
A minimal chat completion against a small model is enough to confirm auth + balance. If you get 401, check key copy/paste. If you get 402, the workspace balance is insufficient or still settling — refresh Billing and retry.
For Claude Code specifically, see the Claude Code setup guide.
Rate limits and first-call tips
Limits cover RPM (requests/min) and TPM (tokens/min) and scale with plan/tier. On a brand-new key, start with one small payload before enabling a full agent toolchain — parallel first-call bursts are the most common source of HTTP 429. Retry with 1s → 2s → 4s backoff (at most three attempts).
If you see 400, the error names the field: model id, role ordering, max_tokens, or cross-model params (thinking, reasoning_effort) that the chosen model does not accept.
FAQ
Does Claude API credit expire? No. Workspace balance never expires.
Is there a monthly fee? No. Pay only for tokens you use (plus the service markup).
Why did checkout charge more than the credit? Card processing (3.9% + $0.30) is billed as a separate line item. The credit that lands in the wallet is the top-up amount itself.
Can I use Claude models without a separate Anthropic account? Yes — via CodeGateway you call Claude-compatible endpoints with one key and one wallet. You still follow upstream usage policies.
Where do I see spend? Dashboard → Usage / Invoices. Each request logs model, tokens, and settled cost.