Rerank API charges usually appear in a RAG bill alongside embedding and generation models. Few teams top up a rerank account on its own, yet the problems are common: money added to a test key, RMB sent to an international site, the wrong invoice header, or a balance discovered only after it hits zero.
This guide separates vendor top-up from gateway top-up and covers account types, currencies and exchange rates, invoices, balance alerts, refunds, and the procurement process. Prices and available models follow the in-site model pricing page.
Note: Cohere, Jina AI, and Voyage AI are trademarks of their respective owners and are used here only to explain technical and procurement processes. Top-up rules, invoice policies, refund terms, and billing methods may change; follow the official agreement and billing page before acting.
1. First confirm which account the money goes into
| Account type | Typical form | Settlement currency | Suitable scenario |
|---|---|---|---|
| Vendor account (international site) | Prepaid balance or usage-based postpaid | USD and others | Direct integration, single vendor |
| Vendor account (domestic site) | RMB prepaid, bank transfer | RMB | Domestic entity, domestic invoices required |
| AI gateway account | RMB prepaid balance | RMB | One balance and invoice across models |
| Cloud marketplace account | Billed through the cloud account | RMB or USD | Already using that cloud, consolidated settlement |
Balances on different sites and in different regions are usually not interchangeable, and neither are keys. Rerank often shares a balance with embedding, so check the billing items before topping up.
A gateway balance is a prepayment to the gateway account, not a vendor balance; pointing a vendor key at a gateway address will fail. For enterprise procurement, verify the contracting, receiving, and invoicing entities first.
2. Vendor top-up and payment flow
- Register an account and complete email verification; for enterprise use, complete enterprise verification first.
- Confirm whether you are using a trial key or a production key. Trial keys have rate limits and are not suitable for production.
- On the billing page, choose a payment method: prepaid balance, a token package, or usage-based settlement after binding a payment method.
- Confirm the settlement currency, billing address, and tax information; for cross-border payment, also confirm the receiving entity.
- Start with a small payment and confirm that it arrives and that charges are deducted correctly.
- Set a balance threshold alert; platforms that support auto top-up can also set a trigger amount.
- Send one minimal rerank request with the production key to confirm that it works.
- Save top-up records, bills, and payment receipts.
The main differences between vendors are in billing cadence:
- Cohere bills by search unit; trial and production keys have different rate limits, so confirm the payment method before upgrading.
- Voyage bills per token, may offer a free allowance, and settles usage beyond it.
- Jina bills per input token, sells token packages, and supports automatic top-up at a balance threshold.
Postpaid accounts are invoiced monthly and require a stable payment method; prepaid accounts pay first and use later, which suits teams with a capped budget. Minimum amounts and credit reviews follow the page.
3. Gateway top-up flow
Gateway top-up usually involves fewer steps:
- Register at ZSQH AI Gateway and sign in to the console.
- Open the top-up page, choose an amount, and pay online in RMB.
- After the payment arrives, confirm the amount and currency on the balance page.
- Create an API key and point
base_urlat the gateway address. - Send one minimal rerank request to verify.
Before topping up, confirm whether the rerank model is available, whether the balance is shared across models, and how invoices are issued. The model list and prices follow model pricing.
For teams using embedding, reranking, and generation models together, a gateway provides one balance, one set of keys, and one invoice, which simplifies reconciliation. Before integrating, confirm rate limits, timeouts, and data terms.
4. Currencies, exchange rates, and fees
| Cost item | Vendor international account | RMB gateway | What to confirm |
|---|---|---|---|
| Pricing currency | USD and others | RMB | The currency shown on the payment page |
| Exchange rate | Card network or bank rate | Platform RMB pricing | Whether conversion cost is included |
| Cross-border fee | Possible | Generally no cross-border transfer step | Bill details |
| Refund | Returned via the original route | Per gateway rules | Who bears exchange losses and fees |
Paying a USD bill with an RMB card may incur a currency conversion fee; when a refund is returned via the original route, the payer usually bears the exchange loss. Large top-ups can be split into several payments, but do not let the business run dry while waiting for a favorable rate. Leave a buffer in the budget, with the ratio set by finance policy.
5. How invoices are issued
| Invoice basis | Suitable scenario | Note |
|---|---|---|
| Amount topped up | Top up first, reimburse later | Unused amounts are also invoiced; refunds may require a credit note |
| Amount consumed | Monthly reconciliation, book actual cost | Requires periodic requests and a longer process |
Before invoicing, confirm the header, tax number, item description, tax rate, and invoice type; the item description follows the invoicing page. The contracting, paying, and invoicing entities should match, or reimbursement and audits can stall.
Whether an overseas vendor can issue a Chinese VAT invoice depends on the contracting entity and its invoicing rules; have finance confirm before paying. A gateway follows its own invoicing rules, and consolidated invoicing is simpler, but confirm that the item description covers the actual purchase.
6. Balance alerts and usage monitoring
Do not rely on the console page alone. Four layers of monitoring are advisable:
- Balance threshold alert: notify by SMS or enterprise chat when the balance falls below a preset amount.
- Daily spend spike alert: trigger when a day's spend exceeds a multiple of the trailing seven-day average.
- Project-level reconciliation: split billing by key, project, and model to find the fastest-growing calls.
- Failed-request monitoring: distinguish arrears, rate limiting, and server errors so that every 429 is not mistaken for a balance problem.
For reranking, also watch for sudden increases in candidate document counts and background reindexing jobs; both can raise daily spend noticeably.
You can estimate how long the balance will last from an exported daily usage file:
import csv
# Adjust to the actual export columns; cost_cny is the daily spend
daily_cost = []
with open("rerank_usage.csv", newline="", encoding="utf-8") as f:
for row in csv.DictReader(f):
daily_cost.append(float(row["cost_cny"]))
avg = sum(daily_cost[-7:]) / min(7, len(daily_cost))
balance = float(input("Current balance (CNY): "))
print(f"Average daily spend over 7 days: {avg:.2f} CNY/day")
print(f"Estimated days remaining: {balance / avg:.1f}" if avg else "Insufficient usage to estimate")
The script only does arithmetic; actual deductions also depend on taxes, voucher offsets, and refunds.
7. Refunds, validity periods, and vouchers
Read the top-up agreement before paying. Whether refunds are supported, the conditions, and the settlement period are governed by the agreement and written support responses; do not assume that an unused balance can always be returned.
Three cautions about vouchers and free credits:
- They have a scope, and some models may not be eligible.
- They have an expiry date and become invalid afterwards.
- They are not counted as paid balance and do not participate in refunds.
Build the enterprise budget on cash top-ups and treat vouchers as a buffer. Before a large top-up, get the refund terms confirmed in writing.
8. Vendor top-up vs. gateway top-up
| Comparison | Vendor account | ZSQH AI Gateway |
|---|---|---|
| Currency | USD or RMB, depending on the site | RMB |
| Payment method | Prepaid, token package, or bound payment method | In-site top-up page |
| Invoice | Per vendor rules; overseas entities require finance confirmation | Per gateway invoicing rules |
| Model range | Single vendor | One balance across models |
| Balance | Not shared across sites | One account balance |
| Suitable scenario | Single vendor, high volume, direct connection | Multiple models, consolidated procurement and reconciliation |
If you use one vendor at high volume with simple finance, a vendor account is the most direct option. If you use several models and want RMB payment and consolidated invoicing, a gateway is simpler. The two paths can coexist.
9. Enterprise procurement checklist
- Confirm that the using, contracting, paying, and invoicing entities match.
- Confirm the account site, region, and settlement currency.
- Confirm the top-up amount, payment path, and arrival time.
- Confirm the invoice type, header, tax number, item description, and invoicing basis.
- Confirm refund conditions, balance validity, and voucher rules.
- Confirm that the rerank model is available and that rate limits meet concurrency needs.
- Confirm balance alerts, daily budgets, and owners.
- Confirm key permissions, log retention, and data terms.
- Allow working days for cross-month procurement.
- Save top-up records, bills, invoices, and contracts for quarterly audits.
FAQ
Does Rerank API need a separate top-up?
In most cases, no. Rerank and embedding usually share an account balance, with each call billed separately; a gateway balance can also cover multiple models. Follow the billing page.
Can money added to a gateway be refunded?
Follow the gateway's top-up agreement and written support response. Treatment may differ by payment method and amount; confirm refund terms before a large top-up and do not assume a refund is always available.
Why does the account still report insufficient balance after a top-up?
First check whether the key belongs to the account that was topped up, whether the site is correct, and whether the funds have arrived. Then confirm that the model is covered by the balance and whether the error means arrears or rate limiting.
Can an overseas vendor issue a Chinese VAT invoice?
It depends on the contracting entity and the vendor's invoicing rules. Have finance confirm the invoice type, item description, and tax rate before paying; if a domestic VAT special invoice is required, prefer a domestic entity or gateway that can issue one.
How should exchange-rate swings be budgeted?
Leave a buffer per finance policy and review actual deductions quarterly. Teams that need a fixed RMB cost can prefer an RMB gateway, but its pricing may include service cost and should still be compared with a vendor account.
Next step
Register at ZSQH AI Gateway, review the currency, invoicing rules, and available models on the top-up page, then make a small top-up and run a rerank request. Prices, billing units, and the model list follow model pricing.
Related reading:
