OpenAI releases GPT-Live voice models and updates ChatGPT Voice

In short: OpenAI has released two new voice models, GPT-Live-1 and GPT-Live-1 mini, and is switching ChatGPT Voice to use the smaller one by default.

What happened

OpenAI announced new conversational voice models called GPT-Live-1 and GPT-Live-1 mini. The company says they sound more natural and handle back and forth conversation better, including smoother turn-taking.

A key change is that these models are “full-duplex,” which means they can speak and listen at the same time. That is closer to how a phone call works than how most voice assistants work (where one side talks, then the other side responds). OpenAI says this lets you interrupt naturally and makes features like live translation easier.

OpenAI is also replacing ChatGPT’s current “Advanced Voice Mode” with GPT-Live-1 mini for most users. People on paid plans will be able to use the larger GPT-Live-1 model.

In a press briefing, OpenAI said the older setup used separate parts for speech to text (turning your voice into written words), the main AI model for the reply, and text to speech (turning the reply back into audio). The new setup is meant to reduce awkward moments like the assistant talking over you. OpenAI also said the voice system can pass requests to newer text models, such as GPT-5.5, while keeping the conversation going.

Why it matters

If this works as described, talking to ChatGPT could feel less like leaving voice messages and more like a real conversation, especially during longer chats or when you need to jump in and correct something. It could also improve live translation, although OpenAI showed a demo where Hindi translation still sounded unnatural and had a strong American accent.

Source: TechCrunch AI