OpenAI Launches New ChatGPT Voice Assistant with Real-Time Two-Way Interaction

OpenAI has introduced a new generation of its ChatGPT voice assistant, featuring simultaneous listening and speaking capabilities that enable more natural and fluid conversations. The system, based on two models called GPT-Live-1 and GPT-Live-1 mini, allows users to interrupt, pause, or return to previous topics without the assistant misinterpreting silences as the end of a query. The full version is available to paid subscribers (Go, Plus, and Pro), while the smaller model will serve free users. The rollout began globally on the ChatGPT website and iPhone and Android apps, with availability varying by country and app version.

Unlike previous voice assistants that processed speech sequentially, converting voice to text, generating a response, then reading it aloud, the new models operate with full two-way communication. This technology lets the assistant decide multiple times per second whether to listen, speak, pause, or activate other tools, reducing delays and preserving the natural tone and emotion of speech. OpenAI claims the system better handles background noise, overlapping conversations, and speech rate changes by focusing on the user's voice rather than all ambient sounds. Internal tests showed users preferred the new voice experience during 5 to 10-minute conversations.

The upgraded voice assistant also integrates with the advanced GPT-5.5 model for complex tasks such as web searches or deep analysis, allowing the conversation to continue without long waits. Users can select response modes prioritizing speed or deeper reasoning. Visual information like weather forecasts, stock data, maps, or images can appear alongside spoken replies, and written messages can be sent within the same conversation.

A notable feature is live translation, enabling real-time bilingual conversations. However, OpenAI acknowledges uneven performance across languages, with some accents and phrasing sounding less natural. The system includes safeguards for sensitive topics like self-harm or emotional crises, offering expert-reviewed support resources when appropriate.

Certain features, including screen sharing and live video streaming, are not yet supported in the new mode, and business, organizational, educational environments, temporary chats, personalized chats, and desktop apps are excluded from the initial launch. Users needing unsupported features can revert to the previous advanced voice mode.