The OpenAI API uses prepaid, pay-as-you-go billing: if an account has no available credit, requests fail. For teams in mainland China, the usual blocker is not the model but the top-up: which currency to use, which payment method works, how long credit takes to arrive, whether an invoice is available, and what happens to unused balance.
This article separates the two paths — topping up directly with OpenAI and topping up through a domestic gateway — and gives steps you can follow, ways to verify that credit arrived, and an invoice checklist. Specific amounts, exchange rates, and promotions are subject to the on-site pricing and top-up pages; this article does not hard-code any numbers.
Note: OpenAI, GPT, and ChatGPT are trademarks of their respective owners. This article uses the common process as of October 2026. Payment methods, invoice policies, and prices are subject to the official pages and the on-site model pricing.
1. First, separate an API top-up from a ChatGPT subscription
This is the most common misunderstanding. Subscriptions such as ChatGPT Plus and Pro serve the chat product and are billed monthly; the OpenAI API serves programmatic calls and charges by token usage. The two balances do not interoperate, and no matter how expensive the subscription is, it does not automatically become API credit.
Likewise, credit in an API account usually cannot be used to buy a subscription. Keep the two budgets separate: product subscriptions belong under office software, while API top-ups belong under R&D or business costs.
2. Path A: topping up directly with OpenAI
The official path suits teams that have an overseas entity or overseas payment method and only need OpenAI models. The process is roughly as follows; the actual pages in the console take precedence:
- Sign in to the OpenAI developer platform and confirm the current organization and project.
- Add a payment method in the billing settings and confirm that the card supports international payments.
- Buy prepaid credit, or enable auto-reload and set the trigger threshold.
- Set a budget cap and usage alerts for the project to avoid accidental overspending.
- Create a key on the API Keys page and name it by project or environment.
- Verify the balance, model, and network with one minimal request.
A few points that are easy to overlook:
- OpenAI settles in US dollars; the amount charged in RMB depends on the card issuer's exchange rate and cross-border fees.
- Set a cap on auto-reload; otherwise a traffic spike may trigger repeated charges.
- Billing receipts are usually downloaded from the billing page. Whether a Chinese VAT invoice can be issued depends on the contracting entity and platform policy, so enterprise buyers should confirm in advance.
- Do not use unknown "top-up agents," and never hand your account, password, or verification code to a third party.
3. Path B: topping up through a domestic AI gateway
A domestic gateway differs mainly in payment and billing: RMB payment, RMB billing, and one console for multiple model providers. Taking the ZSQH AI Gateway as an example, the process is:
- Open the sign-up page and complete registration and any required company information.
- Go to the top-up page, choose an amount and payment method, and complete payment.
- Confirm that the credit has arrived, then create an API key in the console.
- Change the Base URL to
https://api.zhongshangqihang.com/v1and enter the gateway key. - Send a minimal request to confirm the model, balance, and network all work.
- Split keys by project and set quota limits and alerts.
A gateway usually provides both an OpenAI-compatible endpoint and an Anthropic-compatible endpoint. The former serves tools such as the OpenAI SDK, LangChain, and Dify; the latter serves tools such as Claude Code. Model names are subject to the on-site model pricing; do not simply copy official model names.
| Dimension | OpenAI direct | Domestic AI gateway |
|---|---|---|
| Settlement currency | USD | RMB |
| Payment methods | International credit cards, etc. | Domestic online payment, corporate transfer, etc. |
| Credit arrival | Usually fast, depending on the payment channel | Online payment is fast; corporate transfer may need manual confirmation |
| Invoices | Per official billing policy | RMB invoices per platform policy |
| Model range | OpenAI models | Multiple model providers through one endpoint |
| Best for | Teams with an overseas payment method | Teams that need RMB billing and invoices |
4. Three things to do once credit arrives
First, check the balance and the transaction history. If payment succeeded but the balance did not change, save the order number and payment receipt before contacting support, and do not pay twice.
Second, verify the path with a minimal request:
export ZSQH_API_KEY="sk-your-key"
curl "https://api.zhongshangqihang.com/v1/chat/completions" \
-H "Authorization: Bearer $ZSQH_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.5",
"messages": [{"role": "user", "content": "Reply with OK only"}],
"max_tokens": 16
}'
A 200 response with content means the key, balance, and model all work. If you get insufficient_quota, troubleshoot in this order: did the balance arrive, does the key have a spending cap, is the model enabled, and are you being rate-limited?
Third, set budgets. Split keys by project and set a monthly cap and alert thresholds for each key. Do not share one key across all business lines; otherwise you cannot trace the source when the bill looks wrong.
5. Invoice and reconciliation checklist
For enterprise top-ups, confirm the following before paying; it is easier than asking afterward:
| Item | What to confirm |
|---|---|
| Invoice type | Standard or special VAT invoice |
| Billing entity | Company name, tax ID, address and phone, bank and account number |
| Invoice amount | Based on the amount paid or the credit received |
| Timing | How long after payment, and whether retroactive invoices are supported |
| Supporting documents | Order number, contract, payment receipt |
| Refund rules | Whether unused balance is refundable; whether promotional credit is refundable |
When reconciling, look at top-up transactions and API usage separately: top-up records explain how money came in, while usage bills explain how money was spent. If the two do not match, the usual causes are promotional credit, refunds, exchange-rate conversion, or bills that cross month boundaries.
6. How much to top up and how often
Do not top up too much at once, and do not wait until the balance hits zero. A simple estimate:
Days of runway ≈ current balance ÷ average daily spend over the last 7 days
Multiply the average daily spend of the last seven days by a business growth factor, then leave one to two weeks of buffer. That gives a reasonably safe top-up rhythm. For a new project, start with a small top-up, verify the integration and model quality, and then add more.
If you use auto-reload, always set a per-transaction cap and a monthly cap. Auto-reload solves "forgetting to top up," not "runaway usage"; the latter requires budgets, alerts, and model routing.
7. Common pitfalls
- Mistaking a ChatGPT subscription for API credit and topping up the wrong place.
- Topping up under the wrong project, so another key still has no credit.
- Paying twice because you did not wait for the credit to arrive.
- Sharing one key across all business lines, making an abnormal bill impossible to trace.
- Writing a key into front-end code or a public repository, which leads to theft.
- Confirming only the "credited amount" without confirming the invoice type and timing.
8. FAQ
Is a top-up the same as a subscription?
No. A subscription is for the ChatGPT product; a top-up is for API calls. The balances do not interoperate, and the billing cycles differ.
Can I use official top-ups and a gateway at the same time?
Yes, but there is no need to split the same business across two billing entities. Pick one primary path and use the other only as a backup or for specific models.
How long does a top-up take to arrive?
Online payment is usually fast; corporate transfer may require manual confirmation. The top-up page and support replies take precedence.
Can unused balance be refunded?
It depends on the platform terms. Unused cash balance, promotional credit, and amounts already consumed follow different rules. Confirm at top-up time rather than asking when you no longer need the service.
Why do requests still report insufficient balance after a top-up?
First confirm which project the key belongs to, whether the project balance arrived, and whether the key has a spending cap. Then confirm whether the requested model is enabled. It may also be a rate limit misread as a balance problem.
Next steps
If you need RMB top-ups, unified billing, and RMB invoices, start by registering for the ZSQH AI Gateway, top up a small amount and verify one request, then split keys by project. Prices, exchange rates, and invoice policies are subject to model pricing.
Related reading:
