GPT-Live 1
Audio modelCard summarizedOur premier model for natural, expressive voice conversations with smooth interruption handling.
Source: GPT-Live System Card (covers GPT-Live-1 and GPT-Live-1 mini in ChatGPT; revised 4 Aug 2026)
Where to call it
- OpenAI API
gpt-live-1
Good for
- Spoken conversations that keep going while a backend looks up information or runs tools
- Handling pauses, interruptions and changes in pace (listens while speaking)
- Voice front ends for an existing agent via client delegation
Overview
GPT-Live 1
Our premier model for natural, expressive voice conversations with smooth interruption handling.
- Max length
- —
- Languages
- —
- Modes
- speech-to-speech, full-duplex conversation
- Released
- 10 Sep 2026
GPT-Live-1 is a full-duplex voice model: it listens and responds continuously instead of waiting for a turn to end, and hands harder work and tool use to a separate backend model. The system card was written for the ChatGPT launch on 8 Jul 2026 (GPT-Live-1 for paid users, GPT-Live-1 mini for free users) and evaluates the models with delegation. On 4 Aug 2026 OpenAI re-ran its safety evaluations after finding they had used a mismatched backend configuration, and corrected several values. The model reached the API on 10 Sep 2026. system card §1, change log; API changelog (10 Sep 2026)
gpt-live-1- Typical price
- $0.05 / min
- Weights
- Closed
- Released
- 10 Sep 2026
- speech-to-speech
- full-duplex conversation
- delegation to a backend model or agent
- function calling
- custom voices