ChatGPT Live could make interacting with AI seem like something from a film.
AI voice assistants have been pursuing the sci-fi aspiration for years, but they still struggle to engage in human-like conversations. Most voice systems require clear cues, distinct pauses, and a few seconds before producing a response. OpenAI is now introducing GPT-Live, a new voice model for ChatGPT Voice aimed at making these interactions feel quicker and less prearranged.
The primary enhancement is what OpenAI refers to as a full-duplex architecture. Simply put, GPT-Live can listen and speak simultaneously. It continuously analyzes user input while generating its own replies, allowing it to determine when to speak, when to pause, when to keep listening, and when to utilize a tool.
What distinguishes GPT-Live from previous voice systems?
OpenAI claims that GPT-Live can navigate the more chaotic aspects of everyday conversations more effectively. Users can interrupt ChatGPT with additional questions, pause to think, request it to slow down, or ask it to mute and listen. It can also provide brief acknowledgments like “mhmm” or “got it,” making the dialogue feel less formal.
Introducing GPT-Live, a next-generation voice model designed for more natural human-AI engagement. Starting today, it will be available in ChatGPT. Make sure to turn the sound on for this one. pic.twitter.com/WzoQFvA5ir— OpenAI (@OpenAI) July 8, 2026
The model is also optimized for noisy settings. The company states that ChatGPT Voice should now better isolate the user’s voice in the presence of background sounds, such as traffic or surrounding conversations.
GPT-Live can also manage more complex requests in the background. If a question requires web searching, deeper analysis, or more intricate task execution, it can delegate that to GPT-5.5 and retrieve the result once it's ready. The spoken conversation can persist while this occurs.
Who will get GPT-Live first?
GPT-Live is being launched globally in ChatGPT on iOS, Android, and the web. GPT-Live-1 will power ChatGPT Voice for Go, Plus, and Pro users, while free users will have access to GPT-Live-1 mini instead.
OpenAI
Additionally, OpenAI is incorporating visual cards into voice chats. Therefore, if you inquire about the weather, stocks, sports, or similar subjects, ChatGPT can display a card on the screen while maintaining the voice conversation. Features such as search, memory, images, and file uploads will also remain operational with ChatGPT Voice.
Currently, the main limitation pertains to video capabilities. At launch, GPT-Live can facilitate voice conversations within ChatGPT but does not yet support video or screen sharing. In other words, you cannot directly point your camera at something or share your screen during a GPT-Live voice chat at this time. OpenAI has indicated that these features will be added in the future.
Other articles
NHTSA requires companies developing autonomous vehicles to address issues related to first responder interference by the end of July.
The NHTSA reports that autonomous vehicles are interfering with emergency situations by obstructing ambulances. It has requested that developers of these vehicles provide solutions by the end of this month.
Google's SynthID watermark has successfully exposed its first notable deepfake: a fabricated image of Mitch McConnell.
Snopes utilized Google's invisible SynthID watermark to disprove a widely circulated AI-generated image of Senator McConnell in a hospital bed. The watermark remained intact even after screenshots were taken.
A £35 billion scheme to construct 14 small nuclear reactors in the UK.
Michał Sołowow's SGE intends to install 14 GE Vernova Hitachi BWRX-300 small modular reactors at three locations in the UK, representing a £35 billion investment in power for the AI era, to be completed by 2034.
A Brown University professor's evidence of widespread cheating using mass AI.
A professor at Brown changed his exam to an in-person format, causing the class average to drop from 96 to 48, transforming a vague concern about AI cheating into an undeniable statistic.
Samsung intends to establish a floating AI data center by 2028.
Samsung Heavy Industries intends to develop a specialized 50MW floating data center by 2028, as onshore AI projects face challenges related to land, water, and community opposition.
HubSpot retracts its decision on utilizing customer data for AI purposes.
HubSpot has abandoned its proposal to consolidate customer data for an AI feature following a four-day wave of criticism, highlighting how sensitive the issue of AI and data collection has become.
ChatGPT Live could make interacting with AI seem like something from a film.
OpenAI has introduced GPT-Live, an innovative voice model for ChatGPT Voice that can simultaneously listen and speak, enhancing the speed and naturalness of AI conversations.