OpenAI Response API

The Responses API is OpenAI’s primary interface for generating model output, designed as the successor to the long-running Chat Completions API. Instead of sending a bare list of chat messages, you send an input (a string or structured items) along with instructions, and the API returns a typed list of output items — text, reasoning, tool calls — rather than a single message blob. It folds what used to require multiple products into one endpoint: built-in tools like web search, file search, and code execution, plus stateful conversations the server can carry for you.

It matters because this is the default surface you will target when building on OpenAI models. The Responses API was built for agentic workloads: a single request can involve the model reasoning, calling tools, and producing a final answer, and the response format makes each step inspectable. Server-side state via previous_response_id means you can continue a conversation without re-sending the entire history yourself.

In practice you call client.responses.create() with model, input, and optional instructions, tools, and max_output_tokens. You read the convenience output_text field for simple cases, or walk the output array when tools are involved. Set stream: true to receive semantic events as they happen, and pass store and previous_response_id when you want OpenAI to manage conversation state.