On July 9, OpenAI officially released the GPT-Live series of voice models, including GPT-Live-1 and GPT-Live-1 mini, rolling out globally to ChatGPT users. GPT-Live-1 is available to Plus, Pro, and Go subscribers, while GPT-Live-1 mini is for Free users, with API access planned soon.
The core upgrade is the full-duplex architecture, enabling the model to listen and speak simultaneously—no longer relying on turn-by-turn response. In conversation, GPT-Live can signal attentiveness with phrases like “mhmm” or “yeah,” engage in quick back-and-forth, or stay quiet when you need a moment to think—closer to a real human conversation, with a sense of conversational “rhythm”.
Architectural innovation: a “talk-while-working” division of labor. When handling tasks requiring web search, deep reasoning, or complex work, GPT‑Live delegates to backend GPT‑5.5 models while the front-end voice model continues the conversation. This decouples voice interaction from deep reasoning into separate components, allowing the reasoning engine to be updated independently without retraining the voice model. At launch, backend models include GPT‑5.5 Instant (GPT‑Live‑1/1 mini), GPT‑5.5 Thinking Medium (GPT‑Live‑1 Medium), and GPT‑5.5 Thinking High (GPT‑Live‑1 High).
In 5-10 minute paired dialogue tests against Advanced Voice Mode, GPT‑Live‑1 showed clear leadership in overall preference, turn-taking, interruption handling, fluency, and naturalness, achieving a 75.7% preference score.
OpenAI disclosed that over 150 million people use ChatGPT Voice and Dictation features weekly, spanning hands-free assistance, language practice, bedtime stories, and commute chats. GPT‑Live also supports visual cards for weather, stocks, and sports, and continues to support search, memory, image, and file uploads.
On safety, OpenAI stated that GPT‑Live underwent audio-native evaluation, synthetic audio testing, and internal red-teaming across areas including self-harm, psychosis, emotional attachment to AI, violence, and sexual content, performing comparable to or better than Advanced Voice Mode across all dimensions.
The launch of GPT‑Live marks a transition from AI voice interfaces that “hear and speak” to ones that converse like real humans. The full-duplex architecture and “talk-while-working” design not only make conversations more natural but also open the door to voice as the operating system interface for AI. With 150 million people talking to AI weekly, this interaction shift is itself a paradigm change.