GPT-Live 1 Voice Model Lands on Vercel's AI Gateway

OpenAI's GPT-Live 1, a full-duplex voice model, is now available through Vercel's AI Gateway, giving developers a hosted route to add it to voice applications alongside a mechanism for handing reasoning off to a separate text model mid-conversation.

GPT-Live 1 can listen and speak at the same time. Many voice models rely on turn detection to decide when to respond; full duplex removes that boundary, so a user can pause, interrupt, or add detail while the model is speaking.

The model supports client delegation: when a session needs deeper reasoning than the voice model alone can provide, the calling application receives a delegation-created event, generates a response with any text model available on AI Gateway, then returns the result tagged with the same delegationId so the voice model can speak it. Delegated results are routed onto a dedicated 'commentary' channel rather than mixed into the main conversation.

Delegated model requests are billed separately through AI Gateway from the underlying voice session, and it is the calling application, not GPT-Live 1 itself, that controls permissions, confirmation, and cancellation for whatever the delegated model does.

Integrating the model today requires a specific stack: AI SDK 7, @ai-sdk/openai 4.0.67 or later, and a raw WebSocket client. Vercel's own example snippets omit connection setup, audio streaming, and shutdown handling, pointing developers to its longer AI Gateway guide for a complete implementation.

The release extends AI Gateway's role from routing text models to also hosting real-time audio models with a delegation pattern that lets a lightweight voice session call out to any text model on the platform without breaking the conversation.