Embedding is rarely sold on its own. It usually sits under the account balance of an LLM platform or AI gateway. "Topping up an Embedding API," therefore, really means funding the account that carries it. Teams in mainland China tend to get stuck on four things: which currency to pay in, how exchange rates and fees are calculated, who is alerted when the balance runs low, and how invoices and refunds are handled.

This article breaks the process into two paths, "official account" and "domestic gateway." Specific exchange rates, prices, and invoice policies follow model pricing and the top-up page. This article does not hard-code any numbers.

Note: OpenAI, Anthropic, DeepSeek, and other names are trademarks of their respective owners. This article uses common workflows as of October 2026 as examples. Payment methods, currencies, and invoice policies may change; verify official pages before paying.

1. Distinguish the three "balances" first

Before topping up, clarify which layer is limiting you:

  1. Platform account balance: the actual available funds. A top-up increases this layer.
  2. API key spending limit: each key under the same account can have its own limit.
  3. Project or organization budget: internal cost control allocated by project.

A common misunderstanding is "if there is money in the account, calls must work." If a key has a spending limit, or a project budget is exhausted, requests still fail. When investigating insufficient balance, first see which layer the error points to, then decide whether to top up or change configuration.

Also distinguish cash balance from promotional credit. Promotional credit usually cannot be refunded or invoiced, may expire, and may be consumed after cash depending on the deduction order. Enterprises should record the two types separately so month-end reconciliation stays clear.

2. Official top-up: for teams with overseas payment capability

The official path usually means binding a bank card in the vendor console, purchasing prepaid credit, or enabling auto-recharge. The advantages are a direct billing entity and early access to new models. The cost is that you need a usable overseas payment method, stable console access, and must handle exchange rates, taxes, and documentation yourself.

Confirm three details before paying:

  • Auto-recharge rules: trigger threshold, amount per recharge, and number of retry attempts. Set the threshold too high and you discover the balance only when it hits zero; set it too low and charges may occur too frequently.
  • Foreign currency conversion fee: the issuing bank or card network may charge a currency conversion fee, making the actual cost higher than the displayed price.
  • Document type: whether the vendor provides a receipt, an invoice, or a local tax invoice, and whether it satisfies domestic finance requirements.

This path suits teams with an overseas entity, a single model vendor, and finance processes that accept foreign-currency documents. If the company needs RMB special invoices or uses several model vendors at once, the gateway path is usually less work.

3. Gateway top-up: RMB settlement and a unified bill

A typical domestic AI gateway flow is: register an account, top up in RMB, create an API key, and call through an OpenAI-compatible address. Embedding and chat models share one balance and one console, and billing is consolidated in one place.

Converting RMB into USD-denominated credit involves three cost layers. Read the top-up page carefully:

  1. Settlement exchange rate: which point in time is used, and whether it is a live rate, end-of-day rate, or fixed rate.
  2. Payment channel fee: bank cards, corporate transfers, and third-party payments have different fee rates.
  3. Service fee: whether a platform service fee is added on top of the exchange rate.

If the page says only "pay this much, get that much" without stating the conversion basis, fees, and crediting details, request a written explanation before paying. Before a large top-up, verify crediting and the call path with a small amount.

Comparison Official account Domestic gateway
Payment method Overseas bank card or foreign-currency channel RMB corporate or personal payment
Billing currency Usually USD RMB orders
Invoices Per vendor policy Usually RMB invoices available
Model coverage Mainly one vendor Multiple vendors through one endpoint
Crediting speed Real-time or per platform rules Online payment is faster; corporate transfer may need manual confirmation
Main risk Payment and access stability Dependence on the gateway's operations and transparency

One reminder: a gateway is an independent third-party service, not a branch of the model vendor. When choosing one, look at operating history, documentation completeness, whether balance and usage can be exported, and how incidents are handled.

4. Steps from top-up to first call

  1. Register an account and complete personal or enterprise verification as required.
  2. Open the top-up page and confirm currency, exchange rate, fees, and expected credited amount.
  3. Pay a small amount first, save the order number, and verify that the balance and transaction record match.
  4. Create a dedicated API key with a spending limit. Do not test with the main account key.
  5. Send one minimal embedding request to verify the key, balance, and model all work.
  6. Configure low-balance alerts and daily usage notifications.
  7. After everything checks out, add the larger amount according to the monthly budget.

The verification request can be sent like this. Use the model names shown on the site's pricing page; do not copy the example blindly:

curl "https://ai.zhongshangqihang.com/v1/embeddings" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"your-embedding-model-name","input":["connectivity test after top-up"]}'

A response containing a vector array means the path works. If you get an authentication error, first check whether the key was copied completely. If the response mentions insufficient balance or quota, investigate in the order "account balance, key limit, project budget" rather than paying again.

5. Balance alerts and reconciliation

Embedding consumption has two peaks: initial indexing and re-embedding. A single full job can consume several months of ordinary query usage, so alerts should be reviewed daily rather than waiting for the monthly bill.

Configure at least four types of reminders:

  • Notify finance and technical owners when the balance falls below a threshold.
  • Alert when daily consumption exceeds the trailing seven-day average by a certain percentage.
  • Review consumption rankings by key or project to find abnormal calls.
  • Use separate keys and separate budgets for indexing and re-embedding jobs so they do not eat into the online query quota.

For reconciliation, check four columns: order amount, amount actually credited, consumption for the period, and ending balance. Keep order numbers, payment receipts, and usage export files. List promotional and cash balances separately; otherwise it is easy to mistake campaign credit for available funds.

6. Refunds, invoices, and enterprise procurement

Refunds: whether unused cash balance is refundable, how much is returned, and where it is returned follow platform terms. Consumed amounts and promotional credit usually are not refundable. Refunds go back through the original payment channel and may take several business days. Confirm the rules before topping up, not after the project ends.

Invoices: enterprise procurement should confirm at least six items:

  • Invoice type: standard VAT invoice or special VAT invoice.
  • Billing entity: company name, tax ID, address and phone, bank and account number.
  • Amount basis: issued on the RMB amount paid or on the credited balance.
  • Issuance timing: how soon after payment it can be issued, and whether retroactive issuance is supported.
  • Supporting documents: order number, contract, and payment receipt.
  • Special cases: how already-issued invoices are handled after a refund.

Procurement process: put vendor onboarding, contracts, purchase orders, payment terms, data and confidentiality clauses, incident response, key lifecycle, and exit mechanisms into one document. In particular, balance handling and data export at exit are awkward to negotiate after the fact.

7. Security and common pitfalls

  • Do not put API keys in front-end code, Git repositories, or chat logs. Store them only in server-side environment variables.
  • Do not trust "top-up agents" that ask for your account password, and do not pay large amounts into personal accounts.
  • Before paying, verify the domain name and payee name and confirm they match the contracting entity.
  • Use one key per project, and revoke keys promptly when staff leave or projects end.
  • Set an upper limit on auto-recharge to avoid continuous charges from abnormal calls.

FAQ

Does an Embedding API need a separate top-up?

Usually not. Embedding and chat models on the same platform share one account balance, so one top-up covers both. A few platforms may sell quota separately; follow official documentation.

Can I call the API immediately after topping up?

Online payments usually credit quickly; corporate transfers may require manual confirmation. After crediting, verify with one minimal request rather than immediately running a full indexing job.

Which is cheaper: paying in RMB or paying in USD directly?

Do not look at the exchange rate alone. Include channel fees, service fees, invoice costs, and finance processing time. During small-scale testing, crediting speed and transparency matter more than the nominal rate.

Can unused balance be refunded?

It depends on platform terms. Cash balance, promotional credit, and consumed amounts follow different rules, and refund timing depends on the payment channel. Enterprises should confirm and document this before paying.

Is the invoice amount based on the amount paid or the credited balance?

Both practices exist depending on the platform. Confirm the basis before signing or topping up, and clarify the invoice reversal process after a refund so finance records stay consistent.

Next steps

If you need RMB top-ups, a unified bill, and RMB invoices while calling Embedding, Claude, DeepSeek, and other models, start by registering for the ZSQH AI Gateway. Top up a small amount, get one embedding request working, and then split keys by project. Specific prices, exchange rates, and invoice rules follow model pricing.

Related reading: