Async Tool Calling in the Responses API

The model keeps working while it waits on a tool result instead of pausing until that tool returns.

Set async to true on a function or custom tool and the model issues the call, then resumes generating without waiting
Your application still runs the tool. OpenAI does not take over execution or manage background jobs
Tool output is passed back in a later request using the original call_id, so it matches the right call
A wait tool pattern lets the model launch several async calls and later choose which results it needs
Supported by GPT-6 Astra and later models, and only for tools your application runs, not hosted built-in tools