Models

Speech to Speech

View as Markdown

The Speech to Speech API enables real-time voice conversations over WebSocket, billed by minute of audio plus a flat fee per text input message. Supports function calling with web search, X search, collections, MCP, and custom functions.


At a glance

Details
ModalitiesText, Audio → Text, Audio
Audio pricing/ minute or / hour
Text Input pricing/ message
Regionus-east-1

Pricing

The Speech to Speech API charges based on audio duration and text events sent without audio.

Details
Audio/ minute or / hour of audio sent or received
Text Inputper conversation.item.create event

What counts as a text input message

Every conversation.item.create event you send from the client is billed at , with two exceptions:

  • function_call_output items (server-requested tool results) are not billed.

  • Items whose content is input_audio or audio are billed by the audio meter instead.

response.create is not a billable event. It only asks the model to produce the next turn; any audio the model generates in that turn is billed under the audio meter above.


Rate Limits

Details
Concurrent sessionsper team
Max session duration120 minutes

Capabilities

  • Function calling

  • Web search

  • X search

  • Collections search

  • Remote MCP tools


Availability

Details
Clusterus-east-1

Documentation


Last updated:September 12, 2026