OpenAI API
by OpenAI
Getting-started guide
The developer doorway to OpenAI's models — build your own apps on the same AI behind ChatGPT, billed by usage.
Current prices: $2.50/$10 per 1M tokens) · See all plans
Last reviewed 2026-07-13 · for OpenAI API
You create an account, generate an API key (a secret credential that identifies your account), and your code sends a request to OpenAI's servers with that key attached. The model's response comes back to your program, which does whatever you wrote it to do with it. You adjust the request and send it again.
The OpenAI API is how developers build their own software on OpenAI's models (text, images, speech-to-text, text-to-speech, embeddings) — the same underlying models as ChatGPT, but called from your own code instead of a chat window. It's for people writing applications: a chatbot, a summarizer, a coding tool. It is not the ChatGPT app, and a ChatGPT subscription does not include API access — they're separate products with separate billing.
There's no real free tier for building. New accounts may get a small one-time trial credit after identity verification; after that it's pay-as-you-go, billed per token — for the text going in and the text coming out, at a per-model rate, with output usually costing more than input. (There's a fringe "free daily tokens in exchange for sharing your traffic for training" program — skip it for anything real; see security.)
Set a hard spend limit in the billing dashboard before your first project. Agent-style or high-volume usage can run up a bill fast; a monthly cap is your safety net.
You'll write (or paste) a little code to make your first call — this page assumes you're comfortable with that. If you're not coding yet, start with the ChatGPT app instead.
The important data point most beginners get wrong: standard API traffic is NOT used to train OpenAI's models by default (unlike the consumer free ChatGPT). The one exception is the opt-in "free tokens for data-sharing" program — that route does use your inputs and outputs for training, so don't use it for anything sensitive or proprietary.
You are responsible for what your app produces and for the data you send. Don't send other people's personal or confidential information through the API without the right to do so, and add your own checks on model output before showing it to users.
Sign up, generate an API key, and set a monthly hard spend limit in the billing dashboard before you write any code.
Why this one: The budget cap is the single thing that prevents a surprise bill; do it first.
Tweak: Also set a lower "soft" alert so you get an email before you hit the ceiling.
Put your key in an environment variable (not in your code), so it never ends up in a repo or a screenshot.
Why this one: A leaked key is someone spending your money; this is the habit that prevents it.
Tweak: Using GitHub? Add your env file to .gitignore immediately.
For learning and testing, choose one of OpenAI's smaller, low-cost models rather than the flagship — it's plenty for a first project and keeps costs near zero.
Why this one: Most simple tasks don't need the most expensive model; starting cheap builds good cost instincts.
Tweak: Move up to a bigger model only when a task actually needs it.
Send a request that asks the model to summarize a short paragraph of text into one sentence, and print the response.
Why this one: A single round-trip shows the whole loop — key, request, tokens billed, response — with almost no cost.
Tweak: Swap the summarize task for something from a project you have in mind.
Want the full picture first? See the OpenAI API overview for pricing, the free tier and related tools.
