OpenAI Launches GPT-Live Voice Model Enabling Real-Time Conversation and Translation

OpenAI introduced GPT-Live, a new voice model that listens and speaks simultaneously, allowing users to interrupt naturally and enabling features like live translation. The company described this as a new generation of voice models that make AI conversations much closer to real human dialogue. GPT-Live will be integrated into ChatGPT's Voice Mode, with two versions released to users: GPT-Live-1 and the lighter GPT-Live-1 mini. The mini version will be the default for free users, while paid subscribers will have access to the more powerful GPT-Live-1.

Previously, ChatGPT's voice mode relied on separate models for speech-to-text, language processing, and text-to-speech. The new models instead send queries to the latest text models, such as GPT-5.5, for search, reasoning, and agent capabilities during the conversation. This integration also allows the voice mode to present some information visually. Eti Alti, head of ChatGPT Voice at OpenAI, said the company envisions voice becoming a central computing interface for complex, long-term agent work, citing impressive use cases seen with Codex and ChatGPT.

During a live demonstration, OpenAI showcased the model's simultaneous translation abilities, translating between Chinese and English in real time. The company stated the model supports the most widely spoken languages but did not specify which ones. The model can also remain silent for extended periods, listening to conversations until prompted to respond.

Separately, OpenAI plans to release GPT-5.6 to the public tomorrow after the U.S. government lifted restrictions on its distribution, which had been imposed due to concerns about the model's powerful capabilities falling into the wrong hands. Until now, GPT-5.6 was only available to a limited group of vetted organizations.