Buy OpenAI API Key: Two Routes and Five Checks Before You Pay

Key takeaways

• The key itself costs nothing on either route. What you pay for is prepaid credit that the key spends — and that credit is where price, region and payment method actually show up.
• Two routes exist: the official platform and an OpenAI-compatible relay. Your code calls both the same way; an OpenAI-compatible client only needs a base URL, an API key and a model id — not the official host.
• Most failed purchases are not about model quality. They fail at the payment step or because nobody said who holds the credit and whether it comes back.
• Five checks separate an auditable purchase from a hopeful one: a live model list, a decomposed price, per-request usage records, a stated policy for failed requests, and a written refund path.
• The first ten minutes after paying tell you more than any comparison table: create a key, call once, reconcile the deduction, then revoke the key and watch it stop working.

What you are actually buying

Search for a place to buy a key and you will find keys given away everywhere. That is not a trick: an API key is a string that points at a balance. The balance is the product.

Once you see the purchase that way, the questions change from "who sells keys" to something more useful:

• Who holds the credit, and can I watch it move after each call?
• What is the price denominated in, and does it survive contact with a real request?
• If the service stops working for me, what happens to credit I already paid in?

Every check below answers one of those. None of them needs access to a sales deck.

Route 1: the official platform

The official route is the shortest path when it works: an account, a payment method the platform accepts, and a key you create in its dashboard.

Where it stops people is almost never the model list. It is the payment step — a card the platform will not take, a billing country that does not match where you actually work, or a top-up size that is awkward for a small project. If that step is smooth for you, the official route is the least surprising choice.

What it does not give you is flexibility across vendors: you are buying one vendor's models, billed by that vendor, with that vendor's console and that vendor's invoices.

Route 2: an OpenAI-compatible key

The second route is a credential from a relay that speaks the OpenAI API shape. The important part for a buyer is that this is not a different integration — it is the same one with a different address.

An OpenAI-compatible client is configured with three things: a base URL, an API key, and a model id. Documentation for these clients says it plainly: the base URL "will not be https://api.openai.com/v1 (that URL is for the official OpenAI API)" when you are pointed at another provider. Your SDK calls, your prompts and your retry logic stay where they are.

That is why relays are used for reasons other than price: one key and one balance for several model families, one place to see usage, and a payment method you can actually complete.

How to check: open your client's provider settings and look for the three fields. If a provider cannot tell you exactly what goes in each of them, that is the first warning sign.

Five checks before you pay

#	Check	What a good answer looks like
1	Is the model list live and public?	Model ids readable without signing in
2	Is the price decomposed?	A multiplier against the official rate, with input, output and cache-read shown separately
3	Can you reconcile one request?	The usage block in a response maps to one row in your own record
4	What happens to a failed request?	A stated policy, testable in a minute
5	Is there a written refund path?	A page that says where to submit, how long it takes, and where the money returns

Check 2 in practice: read one model's rates, find the published multiplier, and do the multiplication yourself against the vendor's official number. Either it reconciles or it does not. A single blended figure has nothing to verify.

Check 4 in practice: on KuaiAPI failed requests are not billed, and that is a published commitment rather than a favour — send a call with a model id that cannot exist and confirm that nothing was deducted.

Check 5 in practice: "contact us" is not a refund policy. KuaiAPI's refund route is a page you can read before registering; the money returns to the original payment method and the stated handling window is three business days.

The first ten minutes after you pay

Pay first, then run these four steps in order. Each one produces evidence for the next.

1. Create a key in your own console and name it something you will recognise in a usage record later.
2. Point one client at the base URL — OpenAI-compatible clients take the /v1 path; Claude Code takes the host on its own, which is the detail people most often get wrong.
3. Send one minimal call and read the envelope, not just the answer: model id, usage block, finish reason.
4. Reconcile that one call, then delete the test key and confirm it stops working.

If all four pass, you have verified the parts of the purchase that can be verified cheaply. If one of them cannot be done at all, you have learned something worth more than a discount.

FAQ

Do I need a credit card to buy an OpenAI API key? On the official route, yes — a payment method the platform accepts. On a relay, it depends on the operator: many accept local payment rails. Read the top-up page of whichever route you pick before assuming.

Is an API key free? The key is free; usage is not. You are buying credit, and the consumer chat product is a separate offer with its own subscription. A page promising a free paid-API key is a page about something else.

What is the difference between a ChatGPT subscription and API credit? The subscription buys a seat in a chat product. API credit buys tokens for your own code. If your work happens inside a client or a script, you want the latter; if you only type into a chat window, the subscription is usually the cheaper shape.

Can one key cover GPT and other model families? On a relay, yes — that is the point of putting several families behind one credential. On the official platform you are buying one vendor's models only. Either way the client configuration is the same three fields.

Next step

Two pages answer most of the remaining questions and neither needs an account. The model plaza lists each group's multiplier and each model's input, output and cache-read rates, so a quote can be decomposed instead of compared. The status page shows how each group is behaving right now, including when it is not behaving well.

For the surrounding decisions, DeepSeek API Key Buy: A Buyer's Checklist Before You Pay walks through the same purchase checks on a different model family, What Is an AI API Aggregator? covers the operator-level questions, and One API Key for All Models shows the client-side setup when one credential has to serve several tools. The refund page is the written path referenced above, and the English documentation covers endpoints and credential configuration.