GPT-Live 1.1 Adds More Accurate Speech to Vercel’s AI Gateway

OpenAI’s GPT-Live 1.1 is now available through Vercel’s AI Gateway, bringing a full-duplex voice model with more accurate speech transcription and more consistent voice and accent across languages. Existing GPT-Live integrations can switch to the new model by changing the model ID.
- GPT-Live 1.1 is now available on Vercel AI Gateway.
- The full-duplex model supports pauses, interruptions, and added detail mid-reply.
- It improves speech transcription and voice consistency across languages.
- Existing GPT-Live integrations need only the new model ID.
- Client delegation lets applications request text-model assistance during voice sessions.
A voice model built for interruptions
GPT-Live 1.1 listens while it speaks, allowing callers to pause, interrupt a response, or add detail before the model has finished. Vercel describes it as a full-duplex voice model, a setup intended for more natural back-and-forth conversations.
Compared with the previous GPT-Live model, GPT-Live 1.1 transcribes speech more accurately and maintains a more consistent accent and voice across languages. The update is available through the same Live endpoint and client delegation system used by the earlier model.
Existing integrations can use the new model ID
Developers with an existing GPT-Live integration only need to replace the model ID with openai/gpt-live-1.1. Vercel’s example uses the AI SDK and an OpenAI client configured with the AI Gateway endpoint, then creates a realtime session through the Live API.
The source also shows a setup without delegation, where the application starts a session and sends instructions such as asking the model to act as a helpful voice assistant. Vercel says the examples leave out the WebSocket connection, event parsing, transcript assembly, audio streaming, and shutdown steps. Its GPT-Live guide contains connection setup, audio streaming, and complete examples.
Delegated work can run while the conversation continues
GPT-Live 1.1 can also ask an application for help during a voice session. Developers enable client delegation through the session instructions, and the model requests assistance by sending a delegation-created event.
The application controls delegated work, including its permissions, confirmations, and cancellation. In Vercel’s example, the application sends the delegated request to openai/gpt-6.2-sol, although the source says any text model on AI Gateway can be used. The returned text is sent back on the commentary channel for GPT-Live 1.1 to speak.
Delegated model requests are billed separately through AI Gateway, while voice-session usage continues during the delegated work. Vercel says AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations for higher-than-provider uptime. It also offers custom reporting, Zero Data Retention support, API key budgets, and routing rules. The service reflects provider pricing with no markup and does not charge a platform fee on inference, including Bring Your Own Key requests.
Vercel: GPT-Live 1.1 now available on AI Gateway


