OpenAI Function Calling Guide

What the guide teaches

Function calling is OpenAI's name for tool calling: the model replies with a request to run one of your functions, your code runs it, and you send the result back. This guide is the official walkthrough for the Responses API, OpenAI's current interface for building assistants and agents.

It covers:

• Defining function tools: each tool has type: "function", a name, a description and a parameters JSON Schema.
• Reading calls: the response output contains function_call items with a call_id, the function name and JSON-encoded arguments.
• Returning results: you append a function_call_output item with the same call_id and the result, then call the model again.
• Tool choice: auto by default, required to force at least one call, a named function to force that one, or none.
• Parallel calls: several calls in one turn, and parallel_tool_calls: false when you want at most one.
• Strict mode: strict: true so arguments always match the schema, which requires additionalProperties: false and every field listed as required.

The guide also covers streaming function calls, tool search for large tool sets, namespaces and custom tools.

Why read it if you already know another provider

Because the concepts transfer almost one to one. Anthropic calls the same things input_schema, tool_use and tool_result; OpenAI calls them parameters, function_call and function_call_output. Seeing two providers side by side is the fastest way to understand what is essential to tool calling and what is just naming, which is exactly the knowledge you need to write a thin adapter or use a gateway that switches models.

Prerequisites and time

Comfort with Python or JavaScript, JSON and making an HTTP or SDK request. Reading the guide and running its examples takes two to three hours. Keep it open as reference while you build.

Free or paid?

Reading is free. Running the examples needs an OpenAI API key, and usage is billed per token, so set a monthly budget in your account before experimenting.

What you can do after it

Build a manual tool loop on the Responses API, write strict schemas that pass validation, handle parallel calls and explain the differences between OpenAI and Anthropic tool calling in an architecture review.

Common questions

Is this the same as the OpenAI Agents SDK? No. The guide is the raw API; the Agents SDK is a framework built on top of it that runs the loop, handoffs and tracing for you.

Does Chat Completions still support function calling? Yes, and the guide shows examples for both interfaces, but it notes that some newer models require the Responses API for tool calling, so start new agents there.