ทันเอไอ

OpenAI Launches GPT-Live, Upgrading ChatGPT Voice for Natural Human-like Conversations

A new family of voice models capable of listening and speaking simultaneously, making AI conversations natural and instantly interactive like never before.

OpenAI announced the launch of GPT-Live, a new generation voice model designed to elevate human-AI conversations to be natural and fluid, as if talking to a real person. This new family of models has begun powering the ChatGPT Voice feature as of July 8, 2026. At the heart of GPT-Live is its full-duplex architecture (full two-way communication), which completely differs from previous half-duplex voice models that operate on a turn-taking basis. This capability allows GPT-Live to listen to user input and speak its response simultaneously, eliminating the need to wait for the user to finish a sentence before processing, thus ensuring seamless conversations. The model can make decisions multiple times per second on whether to continue speaking, continue listening, stop speaking, interrupt when a user interjects, or even invoke other tools as appropriate for the conversation. Behind the scenes, when GPT-Live needs to handle complex commands, such as web searches or deep reasoning, it delegates these tasks to large language models like GPT-5.5 for processing, to ensure accurate and comprehensive answers. For public access, OpenAI has divided the model into two versions: GPT-Live-1, which will become the default voice model for paid users (Go, Plus, and Pro), and GPT-Live-1 mini, which will be available for free users. This launch comes amid rising demand for voice features, with OpenAI revealing that over 150 million users engage with ChatGPT's voice and dictation features weekly. However, in its initial rollout, GPT-Live will not yet support video conversations or screen sharing, but the company states it is rapidly developing these capabilities for future additions.