The short answer
An AI API (application programming interface) is a way for software to send a request to an AI model and get the answer back automatically, without a person typing into a chat app. You send text, files or images with an API key, and pay per token used. OpenAI, Anthropic, Google, DeepSeek and others sell API access separately from their chat plans. OpenAI says data sent to its API has not been used for training since 1 March 2023 unless you opt in.
| What it is | A programmable door to an AI model |
|---|---|
| How you pay | Per token of input and output, not a flat monthly fee |
| What you need | An account on the developer platform and an API key |
| Separate from chat plans | Yes, though Claude Max, Ultra and Team and ChatGPT Pro include monthly API credits |
| Training on your data | OpenAI and Anthropic: not by default for API traffic |
| No-code route | Zapier, Make and n8n call AI models for you |
What an AI API is, in plain English
A chat app such as ChatGPT or Claude is a finished product for people. An API is the same kind of model offered to other software. Instead of typing into a box, a program sends a structured request over the internet: which model to use, the instructions, the user's text, maybe a file. The model's answer comes back as data that the program can put wherever it likes, in a spreadsheet, a support ticket or a product page.
TechCrunch's AI glossary describes API endpoints as "buttons" on a piece of software that other programs can press to make it do things. That is a good mental picture. When your help desk drafts a reply or your CRM summarises a call, it is very likely pressing one of those buttons on a model maker's API.
How an AI API differs from a chat subscription
| Chat app (ChatGPT, Claude, Gemini) | AI API | |
|---|---|---|
| Who uses it | A person, in a chat window | Software, automatically |
| How you pay | Free plan or monthly subscription with usage limits | Pay as you go per token |
| Features | Memory, projects, file creation, voice and other app features | The model plus the tools you build or switch on |
| Control | The app decides the instructions and settings | You write the system instructions and choose the model |
| Training on your data | Depends on plan and privacy settings | OpenAI and Anthropic: not by default |
The two are billed separately. A ChatGPT Plus subscription does not pay for API calls, and API credit does not upgrade your chat plan. There are exceptions worth knowing: Anthropic includes monthly API credits on Claude Max, Ultra and Team, and OpenAI includes monthly API credits on ChatGPT Pro. Prices change often, so check Anthropic's pricing page and OpenAI's pricing page rather than any figure you read elsewhere.
Who offers AI APIs in November 2026
- OpenAI: the Responses API with models such as GPT-6 Astra, which its documentation recommends as the starting point for reasoning work.
- Anthropic: the Claude API with Claude Opus 5.5 (model id claude-opus-5-5), Sonnet 5.5 and Haiku 5.5.
- Google: the Gemini API through Google AI Studio, including Gemini 4 Argon.
- DeepSeek: deepseek-flash and deepseek-v4-pro, with thinking mode on by default.
- Z.ai: GLM-5.3 and smaller models, including free API models GLM-4.7-Flash and GLM-4.5-Flash.
- Mistral AI: Mistral Large 4 in public preview in Mistral Studio since 6 October 2026.
- Meta: the Meta Model API with Muse Spark models and Muse Wonder.
Why it matters at work
You do not need to be a developer for APIs to affect you. Three situations come up again and again:
- Automating repeat work. If you paste the same kind of text into a chat every day, an API call inside an automation can do it for you. Zapier, Make and n8n handle the API side so you only fill in fields. Our guide to automating tasks with AI shows how, and Zapier vs Make helps you pick.
- Privacy reviews. When a vendor says its feature "uses GPT" or "uses Claude", it usually means the API. Ask which provider, which region and whether the provider can train on the data. Anthropic's privacy centre says it does not use inputs or outputs from its commercial products, including the API, for training by default. Our guide on keeping your data private in AI tools has the full checklist.
- Cost surprises. APIs bill per token, and on reasoning models the hidden thinking is billed as output too. OpenAI notes that a request can run out of room and stop before any visible answer, while you still pay for the input and reasoning used. Set a spending limit before you build anything.
An example you can read in two minutes
This is what a request to OpenAI's Responses API looks like, adapted from OpenAI's own text generation guide. You do not need to run it to understand it:
- The address: https://api.openai.com/v1/responses, the button being pressed.
- The key: a header that carries your API key, which works like a password for your account.
- The model: "gpt-6-astra".
- The instructions: for example "Answer as a concise HR adviser." OpenAI says instructions given this way take priority over the prompt itself. This is the API version of a system prompt.
- The input: the user's question, such as "Summarise our parental leave policy in five bullet points."
If you want to try it without code, build a two-step automation in Zapier or Make: a new form entry triggers an AI step that writes a summary, which is then posted to a Slack channel. You will have used an AI API without writing a line of code.
Common misconceptions about AI APIs
- "The API is just ChatGPT without the website." Not quite. Chat apps add their own system prompts, memory and tools. The API gives you the model and leaves the rest to you, so answers can differ.
- "An API key is a setting I can share." It is a secret that lets anyone spend your money. Never paste it into a chat, a shared document or a public code repository.
- "Every model accepts the same settings." They do not. Anthropic has removed the temperature setting on Claude Opus 5.5 and Google says Gemini 4 ignores it. OpenAI's Responses API still lists temperature, but OpenAI says to remove it for GPT-6 models whenever reasoning effort is not none, and GPT-6 Astra has no none setting. See what is temperature in AI.
- "Using the API means my data is private." It means the provider's commercial terms apply. Check retention, region and training terms for each provider and each tool built on top of it.
Related terms
- Token: the unit APIs bill in.
- MCP: a standard way to wrap APIs so AI apps can use them.
- System prompt: the instructions you set in an API request.
- LLM: the model on the other end of the API.
- Inference: what each API call pays for.
- Open-source model: the alternative to calling someone else's API.
For tool-by-tool detail, see our guides to ChatGPT, Claude, Gemini and DeepSeek, or the best AI for automation.
Prompts to copy
Turn a repeat task into an automation plan
Every [day/week] I [describe the repeat task, for example copy new support emails into a spreadsheet and write a one-line summary of each]. Design an automation that does this with an AI API step in [Zapier, Make or n8n]. List the trigger, each step, the exact instructions the AI step should get, the fields it should return and where a human should review the output.
You get: A step-by-step automation you can build without code
Question a vendor about its AI API use
We are reviewing a software vendor whose product uses AI. Write 10 precise questions to send them about which AI provider and model they call, where data is processed, how long prompts and outputs are kept, whether the provider or the vendor can train on our data, and how we can switch the feature off. Keep each question to one sentence.
You get: A ready-to-send due diligence list