OpenAI launches GPT-Live: new voice AI with near-instant response time

OpenAI has unveiled GPT-Live, a new voice AI model built directly for speech rather than routing through text. The model powers ChatGPT Voice and responds with a delay of under 300 milliseconds, making conversation feel far more natural and fluid.

What sets GPT-Live apart is that it removes the so-called text bottleneck: earlier voice systems first had to convert speech into text, process the text, and then turn the answer back into speech. GPT-Live works more directly with audio, which both lowers the response time and allows for greater emotional nuance in the voice.

The launch is another step toward conversations with AI resembling ordinary conversations between people.