deniz.in

Markets

Weather

Loading weather

· via Vercel blog

OpenAI's GPT-Live 1.1 full-duplex voice model arrives on Vercel AI Gateway

OpenAI's GPT-Live 1.1 voice model is now available through Vercel's AI Gateway, adding full-duplex speech that listens while talking, plus easier delegation to text models.

OpenAI's GPT-Live 1.1 full-duplex voice model arrives on Vercel AI Gateway

GPT-Live 1.1 lands on AI Gateway

Vercel has announced that GPT-Live 1.1, OpenAI's full-duplex voice model, is now available through its AI Gateway. According to the Vercel blog, the model listens while it speaks, so a caller can pause, interrupt, or add detail while a reply is still being delivered — a shift away from the strict turn-taking of conventional voice assistants.

Improvements over the previous model

The Vercel blog says GPT-Live 1.1 transcribes speech more accurately than the earlier GPT-Live release and keeps a more consistent accent and voice across languages. For applications that switch languages mid-conversation or serve multilingual users, that consistency matters, since voice character can otherwise drift between responses.

A drop-in upgrade for existing integrations

Because the new model uses the same Live endpoint and client-delegation mechanism as its predecessor, Vercel says an existing GPT-Live integration only needs the new model ID, openai/gpt-live-1.1, to switch over. Developers can also try the model directly in the AI Gateway model playground.

Delegating work to text models

Client delegation lets a voice session hand work off to other models while the conversation continues. Developers can use session instructions to tell GPT-Live 1.1 when it should ask for help; when it does, it emits a delegation-created event. The application can then answer that request with any text model available on AI Gateway — Vercel's example uses openai/gpt-6.2-sol — and return the result on the commentary channel for GPT-Live to speak aloud.

Vercel notes that the application stays in control of delegated work, including permissions, confirmations, and cancellation. Delegated model requests are billed separately through AI Gateway, and voice-session usage continues to accrue while those requests run. The published code snippets omit the underlying plumbing — WebSocket connection setup, event parsing, transcript assembly, audio streaming, and shutdown — which Vercel covers in its separate GPT-Live guide.

What the gateway provides

AI Gateway is Vercel's unified API for calling models from multiple providers, tracking usage and cost, and configuring retries, failover, and performance optimizations intended to deliver higher uptime than connecting to a provider directly. It also includes built-in custom reporting, Zero Data Retention support, budgets for API keys, and routing rules. Vercel says the gateway mirrors provider pricing with no markup and charges no platform fee on inference, including on Bring Your Own Key requests.

Why it matters

Full-duplex voice changes how voice applications feel. Users can interrupt or add information mid-reply rather than waiting out a scripted response, which brings the interaction closer to natural conversation. Making that capability available through a gateway with unified billing, failover, and observability lowers the operational barrier for teams that already route model traffic through Vercel.

The delegation mechanism may be the more significant part of the announcement. It lets a real-time voice session pull in a separate text model for specific tasks — with the application mediating permissions, confirmations, and cancellation — without ending the voice session. That pattern points toward voice interfaces acting as conversational front ends to agentic workflows, where the model that speaks and the model that reasons need not be the same system.

The drop-in nature of the upgrade also matters for anyone already running GPT-Live in production: changing a model ID is a low-risk path to better transcription and more stable multilingual voice output.

  • #openai
  • #vercel
  • #voice-model
  • #ai-gateway
  • #real-time-voice

Related posts