OpenAI launches GPT-Live voice models for ChatGPT

OpenAI on Wednesday released GPT-Live-1 and GPT-Live-1 mini, a pair of full-duplex voice models that allow ChatGPT to listen and speak simultaneously — a departure from the turn-based interactions that have defined AI voice assistants. The models are rolling out globally starting today.cnet+1

A Voice Model That Knows When to Be Quiet

The new GPT-Live architecture addresses one of the most common frustrations with voice AI: the inability to handle natural conversational overlap. Rather than forcing users into a rigid call-and-response pattern, GPT-Live-1 can absorb interruptions, issue brief acknowledgments, and continue speaking without losing context.openthemagazine+1

"GPT-Live can still remain in conversation with the user," said Kundan Kumar, research lead for the GPT-Live model, describing it as OpenAI's "smartest voice model" yet. Kumar, a voice AI researcher who joined OpenAI from Character.AI, has led the company's audio model development since early 2026.siliconangle+2

Routing and Visual Supplements

When GPT-Live-1 encounters queries that demand deeper reasoning or real-time information, it automatically routes them to OpenAI's text models for processing, including web search capabilities — all while maintaining the voice conversation. The upgraded model also supplements discussions about topics like weather, stocks, and sports with AI-generated visuals showing relevant information, bridging the gap between voice-only and multimodal interaction.theverge+2

A Crowded Week for AI Releases

The GPT-Live launch arrives during one of the busiest weeks in AI this year. OpenAI is also preparing to publicly release GPT-5.6 Sol on Thursday, its most advanced text model, after a government-requested delay was lifted. Meta ↗ 3.48% launched its Muse Image generation model this week, and xAI is reportedly readying a new model in partnership with Anysphere, the company behind the coding tool Cursor.bloomberg+2

The GPT-Live models represent the consumer-facing culmination of work that began with OpenAI's developer-focused GPT-Realtime-2 in May, which brought GPT-5-class reasoning to voice interactions via API. With full-duplex audio now reaching ChatGPT's hundreds of millions of users, OpenAI is betting that voice will become a primary interface for AI — not just a convenience feature.reuters+1