OpenAI launches GPT-Live-1 and GPT-Live-1 mini, delivering a full-duplex conversational voice experience designed for natural, fluid interactions.

Users can experience the update by updating the app or waiting for a gradual rollout, with notable changes including the ability to interrupt mid-answer and request a slower pace.

Internal tests show the system can maintain context over 30 to 40 minutes of conversation, even on mobile use like walking or commuting.

Live translation shows limitations, such as an American-accented Hindi output and a formal tone, with no published list of supported languages or quality tiers.

The company provides a lighter free version alongside the main model, without releasing detailed technical gaps beyond model size and cost.

Unified design enhances context handling, reduces AI talking over users, and can pause silently to absorb information for smoother conversations.

For tougher questions requiring web search or deeper reasoning, GPT-Live performs behind-the-scenes handoffs to a frontier model to maintain flow.

Voice responses are integrated without a separate setting, and nine voices have been remastered to retain distinct personalities.

Both models are full-duplex, enabling simultaneous listening and speaking with real-time interruption and features like instant live translation.

Voice is positioned as the future primary computing interface to handle complex tasks through conversation, with over 150 million users already using voice/dictation features in ChatGPT.

Competition includes Apple and Amazon improving voice assistants, with startups like Sesame pursuing extended conversational capabilities; safety features cover age-appropriate responses and sensitive-topic resources.

rollout is active on chatgpt.com and iOS/Android apps, with automatic updates once an account is reached and no user toggle required.