Audio Inputs

Audio input lets you pass spoken audio — voice messages, call recordings, interviews, podcasts — directly to chat models that can reason over speech. The model receives the audio and responds in text, all through the standard POST /v1/chat/completions endpoint.

Format#

Audio inputs use the OpenAI-compatible input_audio content part inside a user message's content array, alongside any text parts:

{
  "role": "user",
  "content": [
    {
      "type": "input_audio",
      "input_audio": {
        "data": "<base64-encoded audio>",
        "format": "wav"
      }
    },
    {
      "type": "text",
      "text": "Transcribe this and summarize the main points in Arabic."
    }
  ]
}

Common formats: wav, mp3, ogg, flac, m4a. Audio is passed inline as base64.

Supported models#

Audio input is available on models with the audio_input capability. Query the catalog with ?capability=audio_input.

Audio inputs vs. the transcription endpoint#

Use audio inputs when you want a chat model to reason over speech (summarize a call, answer questions about a recording) in the same turn as a text instruction. Use the dedicated speech-to-text endpoint when you just need a transcript.

Pricing#

Audio sent to a chat model is billed as part of that model's token usage (audio is tokenized by the provider). The dedicated transcription endpoint is billed per audio second instead — see the pricing reference.

Related#

• Text-to-Speech — generate spoken audio from text
• Speech-to-Text — dedicated transcription endpoint