OpenAI has made GPT-Live-1 available through its API, a model built for real-time voice interaction. The headline feature is full-duplex conversation: instead of the turn-based back-and-forth typical of voice assistants, both participants can speak at the same time, which makes exchanges feel more natural.
The API also claims stronger instruction following, helping developers keep voice sessions on track, plus support for custom voices and telephony. That last piece matters for practical deployment, letting companies plug the model into phone systems and branded voice interfaces rather than only chat-style apps.
This is a single announcement from OpenAI, so there is no independent benchmark data or third-party evaluation to weigh against the company's claims. The capabilities described here come directly from the source material.