Multimodal

Audio

This page is about sending audio into a chat request — an input_audio content part, the way you'd send an image_url part for image understanding. It isn't handled today — no capability gating, no format validation, no duration-based billing. Sending one isn't specially rejected either; it's just not a request shape any model in the catalog is checked against.

This is a different thing from speech-to-text, which is supported — that's a dedicated POST /v1/audio/transcriptions endpoint, not audio inline in a chat message.

What to do today

Transcribe the audio first with speech-to-text, then send the resulting text as a normal chat message.