OpenAI released GPT-Live-1 through its API, giving developers a full-duplex voice model that can listen while it speaks. TestingCatalog independently corroborates the release, simultaneous conversation, interruption handling and twelve voices. OpenAI also adds custom voices and telephony support, prices use at 5 cents a minute, and presents its published performance figures as company measurements rather than independent tests.
OpenAI released GPT-Live-1 to developers through its API. The full-duplex design lets an application listen while it is speaking instead of waiting for one side to finish before the other begins. TestingCatalog independently reports the launch, simultaneous listening and speaking, and interruption handling. The model succeeds GPT-Realtime-2.1, which OpenAI says was already used inside ChatGPT.[1], [2]
The release includes twelve new voices across different accents, dialects and languages, and TestingCatalog also identifies twelve voices in its account. OpenAI adds support for custom voices and telephony, allowing developers to place the model behind phone-based services as well as applications. Yelp is named by OpenAI as an early user for handling reservations made by phone. Usage is priced at 5 cents a minute.[1], [2]
OpenAI reports higher full-duplex interactivity and tool-calling accuracy than the previous model, along with lower turn-taking latency and a higher pass rate on a banking-support test. Those comparisons are the company's own measurements; the retained reporting does not provide an independent benchmark. The launch therefore establishes availability, capabilities and pricing, while real-world latency, accuracy and call quality will depend on later use outside OpenAI's tests.[1]
