How is a spoken answer charged, and can I retry its audio?

The rate is USD 0.010 per 1,000 processed tokens. The requester pays once, and participants share the audio. If speech fails, the written answer remains and no speech tokens are charged.

Reply to the unchanged spoken answer with “Retry audio” to reuse its words and voices without another written-answer charge. Playback and download do not generate the answer again or add usage.

Changing the displayed translation does not translate shared audio. A Narrate request uses the displayed text at the time you select it; the fullscreen reader follows the narration’s language even if the message’s display language changes later.

Scheduled Text and Audio requests can request speech while keeping their saved type. The separate Narrate action keeps its existing behavior.