The company announced that GPT-Live is being launched globally on iOS, Android, and ChatGPT.com, where it will enhance the new ChatGPT Voice experience.
In contrast to earlier iterations of ChatGPT Voice, which waited for users to complete their speech before replying, GPT-Live employs a full-duplex system that facilitates ongoing dialogue. This enables the AI to acknowledge users with quick interjections like “mhmm” or “yeah,” pause during user contemplation, or interrupt organically when needed.
OpenAI noted that this new framework makes conversations feel less mechanical and more akin to chatting with another individual.
For more complex tasks such as web searches, nuanced reasoning, or intricate problem-solving, GPT-Live offloads the heavy lifting to OpenAI’s latest frontier models while still engaging in conversation. Initially, these background functions will be managed by GPT-5.5, allowing users to keep conversing instead of waiting for answers in silence.
The company claims that GPT-Live represents its most intelligent voice model to date and will automatically gain access to future upgrades of frontier models.
OpenAI is rolling out two variations of the model—GPT-Live-1 and GPT-Live-1 mini. GPT-Live-1 will serve as the default voice model for Go, Plus, and Pro subscribers, while GPT-Live-1 mini will power ChatGPT Voice for free users.
The enhanced ChatGPT Voice experience also features more informative visual responses. Users can now view information cards covering topics like weather, stocks, and sports while engaging through voice.
OpenAI stated that GPT-Live enhances noise management and is less likely to interrupt users during natural conversational pauses. Users can also instruct the assistant to remain quiet and simply listen if needed.
According to the company, internal assessments indicated that GPT-Live was favored over the existing Advanced Voice Mode in areas such as conversational flow, turn-taking, interruptions, and overall naturalness. It also showed superior performance in scientific reasoning, web searches, and multi-step customer support tasks.
OpenAI confirmed that GPT-Live has been developed with dedicated voice-specific safety protocols, including protections for discussions around self-harm, emotional dependency, violence, and inappropriate content. The model also has measures in place to prevent voice impersonation by limiting responses to OpenAI’s predefined voices.
At its initial launch, GPT-Live will not support voice conversations with video or screen sharing, though OpenAI indicated that these features are in development, and existing voice modes will continue to offer such functionalities in the interim.
