What is an API?

What it means at work

The chat window is one way to use an AI model; the API is another. When a developer builds AI into a tool, such as a reporting app that writes summaries, that tool talks to the model through the API and the company pays per use rather than a flat subscription. The same idea connects spreadsheets, email and other systems to each other.

What the sources say

• What it is: the Claude API is a RESTful API that provides programmatic access to Claude models. The Claude API overview lists an API key from the Console as a prerequisite. Its Messages API is for sending messages to Claude for conversational interactions.
• Models choosing APIs: the Toolformer paper introduces a model trained to decide which APIs to call, when to call them, what arguments to pass.
• How it is billed: OpenAI quotes API prices per 1M tokens. OpenAI’s pricing page lists gpt-6.1-terra at $1.00 per million input tokens and $5.00 per million output tokens. Rates change, so check the current page.
• Separate from subscriptions: the Codex pricing page from OpenAI states that API token prices are separate from subscription usage.
• Cost control: Anthropic offers a way to count tokens in a message before sending to manage costs. Anthropic’s Claude API also has a batch endpoint with a 50 percent cost reduction for asynchronous processing.

An example

Your finance system exports invoices, and a small script sends each one to an AI model through its API to extract the vendor and due date. You never open a chat; the result appears in a spreadsheet.

A common misconception

Many people think using an API means being a programmer. Increasingly, tools like Claude Code can write the connecting script from a plain-English description. What you still own is the judgment: what data is sent, where it goes, and whether the output is right.

Sources

• Anthropic, Claude API overview: “provides programmatic access to Claude models”
• Anthropic, Claude API overview: “An API key from the Console”
• Anthropic, Claude API overview: “Send messages to Claude for conversational interactions”
• arXiv, Schick et al., Toolformer: Language Models Can Teach Themselves to Use Tools: “a model trained to decide which APIs to call, when to call them, what arguments to pass”
• OpenAI, API pricing: “Prices per 1M tokens”
• OpenAI, API pricing: “gpt-6.1-terra $1.00 $5.00 $0.05”
• OpenAI, Codex pricing: “API token prices are separate from subscription usage”
• Anthropic, Claude API overview: “Count tokens in a message before sending to manage costs”
• Anthropic, Claude API overview: “Process large volumes of Messages requests asynchronously with a 50 percent cost reduction”