How to Get an OpenAI API Key (2026): Create, Fund, Test and Secure It

The error messages on this page are real responses from the OpenAI API, recorded on October 15, 2026.

API key vs ChatGPT subscription

An API key is for software: your own code, scripts and developer tools calling OpenAI models directly, billed per token from prepaid credits on the OpenAI Platform.

A ChatGPT plan (Free, Go, Plus, Pro, Business) is for the ChatGPT apps. It also covers Codex when you sign in with ChatGPT. It doesn't give you API credits, and the two are billed separately.

If all you want is Codex, a ChatGPT Plus plan or higher may be simpler than an API key. See Install Codex CLI.

Step 1: Sign in to the OpenAI Platform

Go to platform.openai.com and sign in with the same account you use for ChatGPT, or create one. Your account belongs to an organization, which holds billing and limits, and work is grouped into projects. A default project is created for you.

The API is only available in OpenAI's supported countries. From anywhere else, requests fail with a 403 (see errors).

Step 2: Create the key

1. Open Settings → API keys (platform.openai.com/settings/organization/api-keys).
2. Click Create new secret key.
3. Choose the owner:
   • You: a user-owned project key for your own development.
   • Service account: for servers, CI and anything that shouldn't depend on one person's account.
4. Pick the project the key belongs to. Usage and limits are tracked per project.
5. Set an expiration date. OpenAI recommends expiring keys and rotating them on a schedule. Your admin may also enforce a maximum key lifetime.
6. Click create and copy the key now. You won't be able to see it again.

Step 3: Add credits

Open Settings → Billing and add prepaid credits. Your organization's usage tier then rises automatically as total purchases grow, and the tier sets both your rate limits and your monthly usage cap:

On the Limits page, set an email alert at a dollar amount you'd notice, and a hard spend limit if you want a firm monthly ceiling. For per-model prices, see OpenAI API pricing.

Step 4: Test the key

Put the key in an environment variable, so it stays out of your code and shell history:

export OPENAI_API_KEY="sk-proj-..."        # macOS / Linux
setx OPENAI_API_KEY "sk-proj-..."          # Windows (then open a new terminal)

Then send one request to the Responses API with the cheapest current model:

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-luna","input":"Say hello in five words."}'

A working key returns a JSON response with the model's text in output and token counts in usage. With GPT-6 Luna at $0.10 per million input tokens, the request costs a tiny fraction of a cent.

Here's what we got with a deliberately invalid key:

{
  "error": {
    "message": "Incorrect API key provided: sk-proj-*******************g123. You can find your API key at https://platform.openai.com/account/api-keys.",
    "type": "invalid_request_error",
    "code": "invalid_api_key",
    "param": null
  }
}

And with no key at all:

{"error":{"message":"Missing bearer or basic authentication in header","type":"invalid_request_error","param":null,"code":null}}

The masked key in the first message (sk-proj-***…g123) shows the first and last characters of what OpenAI actually received. Compare it with your key to spot a truncated paste.

Keep the key safe

• Never ship it in client-side code (browser JavaScript, mobile apps) or commit it to a repository. Call OpenAI from your server.
• One key per app and environment. Separate projects for staging and production let you revoke one without breaking the other, and give each its own spend limit.
• Set an expiration and rotate. Create the replacement, deploy it, confirm it works, then revoke the old key.
• Lock it down further with an IP allowlist (organization security settings) and per-project spend limits.
• If a key leaks, revoke it immediately on the API keys page. Usage can pile up fast.

Common errors

For any 429, read error.code first. Billing and quota errors look similar to rate limits, but they need a top-up or a higher limit, not a retry.

Use the key in tools

• OpenAI SDKs read OPENAI_API_KEY automatically.
• Codex CLI: printenv OPENAI_API_KEY | codex login --with-api-key. Usage is then billed at API prices. See Install Codex CLI.
• Desktop apps such as Cherry Studio: paste the key into the OpenAI provider.

Can't get a key?

If you're outside the supported countries, or can't add a payment method OpenAI accepts, the usual alternatives are Azure OpenAI (billed through an Azure account) or an API gateway that resells model access with its own keys. api.foo, which publishes these docs, is one of them. It accepts OpenAI-format requests at https://api.foo/v1. Whichever you choose, check how it handles your data and billing.

Related

• OpenAI API pricing (with calculator)
• Install Codex CLI and set up config.toml
• How to get a Claude API key