OpenAI on Thursday launched GPT-Live-1 in the API, bringing the full-duplex voice model already used inside ChatGPT to developers building voice apps and phone agents, according to an OpenAI product post. The company said the model can listen and speak at the same time, handle interruptions in a single speech stack, and delegate deeper reasoning or tool calls to a backend such as GPT-6 Astra or a third-party model so conversation can continue while work runs in the background.
Traditional voice agents chain speech-to-text, a reasoning model, and text-to-speech; OpenAI said GPT-Live-1 collapses the voice layer into one model and supports telephony, twelve new voices, ASR transcripts, and response text out of the box. Early customers cited include Yelp Host, Speak, Intercom’s Fin, and Cognition’s Devin. Speak reported nearly 80% fewer interruptions versus prior turn-based tutors, and OpenAI said Full Duplex Bench scores rose about 30 percentage points over GPT-Realtime-2.1. Pricing for the front-end voice layer is $0.05 per minute, with backend model and agent fees billed separately.
THE DECODER independently covered the launch, noting Yelp CTO Alex Levy’s comments on improved call handling for reservations and orders. This brief covers the September 10 API availability and architecture; it does not benchmark production latency or audit any customer’s telephony stack.