GPT-Audio-1.5
GPT-Audio-1.5 works with audio through Chat Completions. It supports text and audio input and can return either text or generated speech.
Audio does not always require a Realtime session
This model uses the Chat Completions REST endpoint rather than the Realtime endpoint. It supports streaming and function calls, with a 128,000-token window. Audio and text have separate token rates.GPT-Audio-1.5 official model documentation
Choose the interaction pattern first
Practical guidance. For a voice note that receives one spoken answer, a request-response flow may be sufficient. Review the recording, response and any tool result as one task. A full duplex conversation requires different interruption and session behavior.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | gpt-audio-1.5 |
| Context window | 128,000 tokens Published model context; input and output share this budget GPT-Audio-1.5 official model documentation |
| Maximum output | 16,384 tokens Published maximum output; bounded by the remaining context GPT-Audio-1.5 official model documentation |
| Input and output | text, audio in; text, audio outGPT-Audio-1.5 official model documentation |
| API routes | Chat Completions (v1/chat/completions)GPT-Audio-1.5 official model documentation |