How OpenAI Developers launched GPT-Live-1 on X

Developers building voice applications on GPT-Live-1 can now create a voice from a text prompt that specifies qualities such as pitch, timbre, texture, accent, and speaking rate, or choose from nine new preset voices. Those presets were created with native-speaking actors in French, German, Hindi, Japanese, and Korean. GPT-Live-1 is the full-duplex voice model in the OpenAI API that listens and speaks at the same time and can hand reasoning and tool calls to a separate backend. The update gives product teams a practical way to match a voice to an app’s role and to listeners who expect a native accent, without supplying a sample recording. Prompt-to-voice is available to all API customers in the Playground and via the API. Voice creation is free, each result is saved under an ID for reuse, and spoken output is billed at standard session rates. Organizations may store up to 100 prompt-generated voices. A launch demonstration applies the same method to a bright shopping assistant, a patient education guide, and a deadpan game character. Adult voices are required, and the usage policies prohibit impersonating a real person.