Microsoft-backed (NASDAQ:MSFT) OpenAI has unveiled GPT-Live, a new generation of voice models designed to power real-time, natural conversations on ChatGPT Voice.
Unlike traditional systems that wait for a user to finish speaking, GPT-Live listens and responds simultaneously. This allows it to handle instant interruptions, recognize pauses and provide brief conversational cues like “yeah” or “mhmm” for a human-like flow.
The system also integrates directly with OpenAI’s broader ecosystem. While handling standard voice interactions natively, GPT-Live automatically routes complex requests—such as web searches or advanced reasoning—to a separate frontier model before delivering the results seamlessly back through the audio interface.
GPT-Live will initially rely on GPT-5.5 for heavier workloads and update as newer models become available.
The company also noted that internal testing showed improvements over its previous Advanced Voice Mode across measures including conversational flow, interruption handling, and user preference during five- to 10-minute conversations.
The update expands ChatGPT Voice beyond audio responses. OpenAI said the feature will begin showing visual cards during voice conversations for topics such as weather, stocks, and sports, while maintaining access to tools including search, memory, image generation, and file uploads.
The new models include additional voice safety measures, including protections against unauthorized voice imitation and systems designed to handle sensitive conversations, using a limited set of approved voices rather than allowing users to create direct replicas of real people.
The company acknowledged that performance may vary across languages, with some languages still experiencing differences in accent accuracy and fluency.
"We’ve optimized GPT‑Live for some of the most popular languages in ChatGPT. For certain languages, the model may have a non-native accent or gaps in fluency. We’re actively working to improve the experience across languages," the company said.
AI Voice: The Next Major Interface
The release marks another step in the race among AI developers, as companies compete to move beyond text-based chatbots and into more persistent, conversational assistants.
Google has been expanding its Gemini-powered voice capabilities, Amazon is rebuilding Alexa around generative AI, and Apple is working to bring more advanced AI features into Siri through its Apple Intelligence platform.
At the same time, startups including ElevenLabs are making a deeper push into AI-generated speech, voice agents, and conversational applications. In February, the company announced its most advanced text-to-speech model, Eleven v3.
Photo: Shutterstock
Login to comment