OpenAI launched GPT-Live on July 8, calling it a new generation of voice models for natural human-AI interaction.
The core change is full-duplex voice. OpenAI says GPT-Live can listen and speak at the same time instead of waiting for rigid turns. It can acknowledge the user, pause, interrupt, stay quiet, and make interaction decisions continuously during a conversation.
OpenAI is rolling out GPT-Live-1 and GPT-Live-1 mini to ChatGPT users globally. GPT-Live-1 will become the default model for ChatGPT Voice on Go, Plus, and Pro plans. GPT-Live-1 mini will become the default for Free users. OpenAI says API access is planned but not yet the launch path.
Voice becomes a front end for background work
GPT-Live is not only a smoother speech model. OpenAI says it can delegate harder tasks to frontier models behind the scenes while keeping the conversation moving.
At launch, GPT-Live uses GPT-5.5 in the background when a question requires web search, deeper reasoning, or more complex work. OpenAI says GPT-Live-1 Instant and GPT-Live-1 mini use GPT-5.5 Instant in the background, while Medium and High use GPT-5.5 Thinking with the corresponding reasoning effort.
That makes the voice layer a coordinator. The user hears a live conversation, while search, reasoning, and more agentic work can happen behind it.
The launch also adds practical ChatGPT Voice changes: remastered voices, better handling of pauses and background noise, visual cards for some answers, and continued support for search, memory, images, and file uploads. OpenAI says voice with video or screen sharing is not supported at launch.
The safety problem is real-time, not just content-level
OpenAI pairs the product launch with a GPT-Live system card. The product post says safeguards can act while the model is speaking, including steering output, surfacing resources, or ending higher-risk conversations.
That matters because voice systems do not fail like text systems. A text answer can be reviewed before it is acted on. A live voice agent can interrupt, comfort, persuade, or escalate in real time.
OpenAI says it expanded audio-native safety testing, red-teamed risks unique to voice, added teen protections, and will monitor emotional reliance after launch. Those are necessary claims to watch as GPT-Live moves from everyday chats into more agentic work.
