OpenAIGPT Live
GPT-Live 1
GPT-Live 1 is a conversational voice model that can listen while it speaks. It can delegate reasoning and tools to a separate backend agent.
The voice session and backend work have separate costs
Live sessions cost $0.05 per minute, measured by the second without rounding to a whole minute. Backend model calls and tools are charged separately. The dedicated Live endpoint accepts text and audio, but not images or video.GPT-Live 1 official model documentation
Design for interruptions and unfinished turns
Practical guidance. A voice interface should stop or adjust its speech when the person interrupts. Keep the result of a delegated action separate from the spoken acknowledgment. Measure whether interruptions preserve the intended task and avoid repeated actions.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | gpt-live-1 |
| Input and output | audio, text in; audio, text outGPT-Live 1 official model documentation |
| API routes | Live (v1/live/sessions)GPT-Live 1 official model documentation |