OpenAI rolls out full-duplex GPT-Live voice to ChatGPT consumer users

OpenAI began rolling out a rebuilt voice mode to ChatGPT consumer users on Wednesday, replacing Advanced Voice Mode with two models, GPT-Live-1 and GPT-Live-1 mini, that can hear a user and respond at the same time.

The update will be made available to free and paid consumer accounts and comes one day prior to the launch of the GPT-5.6 model family by the company.

The main reason behind the shift to GPT Live is architectural. While previous voice models would wait for the human speaker to finish their sentence, GPT-Live operates based on OpenAI’s full-duplex architecture, which involves processing input speech while generating output speech at the same time.

According to OpenAI’s announcement, the model decides many times per second whether to talk, keep listening, pause, cut in, or reach for a tool. In practice, that means a user can talk over it, stop to think without being interrupted, or hear it drop in an “mhmm” to signal it is following along.

OpenAI brings GPT-Live to ChatGPT Voice

Free ChatGPT accounts now default to GPT-Live-1 mini, while subscribers on the Go, Plus, and Pro tiers get the larger GPT-Live-1.

The rollout covers iOS, Android, and the web, and because ChatGPT already works through CarPlay, 9to5Mac noted the update also lands in the car. OpenAI says the nine voices in ChatGPT have been remastered for the new models.

It also said it built GPT-Live to be a conversationalist, not a reasoning engine. For more complex requests, GPT-Live does not handle everything on its own. According to OpenAI, the voice model is capable of delegating tasks like web search, reasoning, and even multistep answers to GPT-5.5 without pausing the voice conversation.

This is supposed to address one of the major shortcomings of ChatGPT’s earlier voice mode, where it sometimes seemed to be less powerful compared to typing. Users will also have the option of selecting between a quick-response mode and a high-effort mode.

OpenAI is also adding near real-time speech translation and visual cards for weather, stock prices, and sports scores, giving users some answers on screen instead of having ChatGPT read everything aloud. The company said more than 150 million people already use ChatGPT features such as Voice and Dictation each week.

GPT-Live improves voice flow but leaves gaps

According to TechCrunch, which covered an OpenAI press conference, the Hindi translation had a strong American accent along with very stiff and bookish language.

Per OpenAI, the model works best for “some of the most common languages in ChatGPT,” although they warn that there might be a problem with the accent or fluency in some languages. Also, OpenAI stated that voice in combination with videos and screen sharing is not supported as of now.

While OpenAI has pitched GPT-Live as a more human version of its entire model lineup, it insists that it is not an AI companion. The company also highlighted safeguards it had put in place to guarantee age-appropriate responses to teens. It also added that crisis resources are in place should a user conversation be geared towards self-harm.

Audio clips from Live and Advanced Voice conversations are retained for 30 days and are not used to train models unless users choose to share them.

Atty Eleti, product lead for ChatGPT Voice, told the briefing he has held 30- to 40-minute conversations with the feature on walks, and argued voice could become a main way people run longer, more complex tasks.

GPT-5.6 launch puts the timing in focus

The voice update is separate from the model release OpenAI has scheduled for Thursday.

According to Cryptopolitan, the approval of GPT-5.6 for public release was granted after the U.S. Department of Commerce had released the embargo imposed during its national security review, and it is available in three versions: Sol, Terra, and Luna.

GPT-Live is set to reach OpenAI’s API after the consumer rollout, with developers able to register for access.

The smartest crypto minds already read our newsletter. Want in? Join them.