OpenAI rolls out new voice models, says ChatGPT can talk more like humans now

OpenAI has launched GPT-Live, the new generation voice model for ChatGPT. The AI startup says that now your conversations with ChatGPT will feel more human. And according to OpenAI CEO Sam Altman, not only is GPT-Live more real, but also magical.

In Short

• OpenAI launches GPT-Live, ChatGPT can now talk more like a human
• ChatGPT can now talk to you while it processes information
• Sam Altman calls it magical and real

Voice assistants have existed for years now, be it ChatGPT or even Apple’s Siri. But for long, conversations with such assistants have felt rather robotic. Often, the AI assistant would sound monotonic, or would pause while it processed information, which would break the natural flow you would have when talking to a human. But now, OpenAI says that its latest generation of voice models, GPT-Live, fixes such issues, making ChatGPT sound more like a human.

The AI startup launched two GPT-Live models – GPT-Live-1 and GPT-Live-1 mini that will replace ChatGPT’s older advanced voice mode. As per the company, these models are better at reducing interruptions when you are speaking, while doing a lot of work in the background while you speak.

That is, instead of you having to wait till ChatGPT processes something, such as translations or research, the AI will continue to speak to you, making the conversation feel more natural.

According to OpenAI CEO Sam Altman, using GPT-Live may finally make him switch from typing to talking to AI. He wrote on X, “GPT-live (next-generation voice) launches today in ChatGPT. It feels magical and ‘real’. I have always preferred typing to talking to an AI, now I think that’s going to shift.”

How does GPT-Live work?

At the centre of GPT-Live is a full-duplex architecture. OpenAI describes this as a way for the AI to listen and speak at the same time. Unlike older voice systems that worked in turns and waited for you to stop speaking before responding, GPT-Live continuously processes information while it talks to you.

OpenAI said this lets the model decide many times each second whether to speak, keep listening, pause, interrupt or use a tool.

In practice, it can use phrases such as “mhmm”, “yeah” or “got it” as you speak, something that a human would generally do.

GPT-Live also separates voice interaction from deeper work. When a query needs search, reasoning or more complex agentic capabilities, the model can hand that task to another system in the background, which at launch is GPT-5.5, and then bring the result back into the conversation when ready.

In a demonstration before the release, GPT-Live carried out simultaneous translation into Hindi while the presenter was still speaking, then joined in with the completed Hindi translation. The updated voice mode can also show visual cards for subjects such as weather, stocks and sports, while continuing to support search, memory, images and file uploads.

How good is GPT-Live?

According to OpenAI, the new models outperformed the older advanced voice mode in both human evaluations and technical benchmarks. In matched conversations lasting five to 10 minutes, GPT-Live-1 and GPT-Live-1 mini were strongly preferred over advanced voice mode, with preferred response rates of 75.7 per cent and 69.2 per cent respectively.

As per the Verge, OpenAI research lead Kundan Kumar called GPT-Live-1 the “smartest voice model” yet from the AI startup.

More than 150 million people use ChatGPT’s Voice and Dictation features each week for everyday help, language practice, bedtime stories and general conversation.

In terms of safety, OpenAI has expanded safety testing for GPT-Live with audio-native and synthetic evaluations focused on areas including self-harm, psychosis and mania, emotional reliance, violence and sexual content. The company said the model includes safeguards that can steer responses towards safer output, surface safety messaging or end a voice conversation in higher-risk cases.

OpenAI has also added protections for teen users, including age-appropriate behaviour and parental controls, and that GPT-Live is designed for conversation rather than voice impersonation, using predefined voices with safeguards against imitating a real person’s voice.

Keep in mind that OpenAI is facing lawsuits over ChatGPT allegedly harming users.

Can I use GPT-Live on ChatGPT?

OpenAI has confirmed that GPT-Live is being rolled out to all users on iOS, Android and the web. GPT-Live-1 is the default model for ChatGPT Go, Plus, and Pro users. On the other hand, GPT-Live-1 mini is the default for free users.

GPT-Live will also be available in the API soon, with developers and enterprises able to sign up for access notifications. It added that the model has been optimised for some of the most popular languages in ChatGPT, though some languages may still show non-native accents or gaps in fluency.

At launch, GPT-Live does not support voice with video or screen sharing inside ChatGPT, and those features will remain available through older voice modes.

- Ends