OpenAI launches GPT-Live to make ChatGPT Voice feel more human

Explore with AI:

OpenAI has introduced GPT-Live, a new generation of voice models that now power ChatGPT Voice. The company says the update is meant to make talking with ChatGPT feel closer to a real conversation, with less stiff turn-taking and smarter answers when a question needs deeper work.

GPT-Live is rolling out globally now across iOS, Android, and ChatGPT.com. Tap the Voice button, and the new experience should start showing up as it reaches your account.

ChatGPT can listen while it talks

Older ChatGPT voice modes still worked more like a walkie-talkie. You spoke, then waited. Advanced Voice Mode improved the sound and speed, but it still relied on discrete turns and often jumped in after a short pause.

GPT-Live is full duplex. It can listen and speak at the same time, and OpenAI says the model can decide many times per second whether to keep listening, speak, pause, interrupt, or call a tool.

In practice, that means you can cut in with a question, stop to think, or ask ChatGPT to slow down. OpenAI also says the model will acknowledge you with short phrases like “mhmm” or “got it,” and it is better at waiting when you pause instead of treating silence as your turn ending. Background noise handling is improved too, so nearby chatter or traffic is less likely to throw it off.

OpenAI remastered ChatGPT’s nine built-in voices for GPT-Live. The system is still limited to those predefined voices, with safeguards against imitating a real person’s voice.

Smarter answers without freezing the chat

OpenAI calls GPT-Live its smartest voice model yet. When a request needs web search, deeper reasoning, or more complex work, GPT-Live can hand that job off in the background and keep the conversation going. At launch, that background model is GPT-5.5.

Users can also pick how hard ChatGPT should think: Instant for speed, or Medium and High when a slower, more careful answer is worth the wait.

Voice chats can now show rich visual cards while you talk, including weather, stocks, sports, and maps. Search, memory, images, and file uploads still work in voice conversations.

OpenAI also says the continuous design helps with live translation and longer, more fluid back-and-forth. In its own head-to-head tests, GPT-Live was preferred over Advanced Voice Mode for conversational flow, and it scored better on several reasoning and search-style evaluations.

Free and paid users get different models

OpenAI is shipping two versions:

• GPT-Live-1: default for ChatGPT Go, Plus, and Pro
• GPT-Live-1 mini: default for Free users

Both are part of the global rollout. Developers and businesses can join a waitlist for API access, which OpenAI says is coming soon.

What is still missing

At launch, GPT-Live does not support voice with video or screen sharing in ChatGPT. OpenAI says those features are coming. Until then, users can still reach legacy Standard and Advanced Voice Mode where those options are available.

OpenAI also notes that GPT-Live is optimized for some of ChatGPT’s most popular languages. In others, users may hear a non-native accent or weaker fluency while language quality improves.

Why this matters

More than 150 million people already talk to ChatGPT each week through Voice and Dictation, according to OpenAI. A full-duplex model is a clear step toward the kind of always-on voice assistant people expect on phones, in the car, and eventually on hardware. That fits the broader direction of a voice-first OpenAI device that leans more on conversation than screens.

The company is also treating voice safety as its own problem, not just a text model with speakers. GPT-Live includes voice-specific safeguards, support flows for sensitive conversations, and teen protections. Parents can control whether a teen can use ChatGPT Voice through parental controls, and linked parents may get alerts in higher-risk self-harm situations.

For most people, the test is simple: open ChatGPT, start a voice chat, and see whether interruptions, pauses, and follow-up questions feel less robotic. If GPT-Live sticks the landing, ChatGPT Voice stops being a novelty mode and becomes a normal way to get real work done out loud.

Explore with AI: