OpenAI Brings Real-Time Voice Model GPT-Live-1 to API with Telephony Support
Developers can now plug low-latency, two-way conversational voice directly into software applications and phone systems.
OpenAI has introduced GPT-Live-1 into its developer platform, providing access to real-time, bidirectional voice conversations over an API. The new release brings native full-duplex speech — meaning you can talk and interrupt the AI naturally without waiting for awkward pauses — straight to custom apps and standard phone lines.
Traditional voice bots string together three separate systems: a speech-to-text transcriber, an underlying language model to generate an answer, and a text-to-speech engine to read it out loud. That pipeline introduces noticeable latency — the slight delay between when you stop talking and when the computer replies. GPT-Live-1 handles audio processing end-to-end as a single system, cutting latency down to human-like conversational speeds.
Telephony Integration and Steerability
Alongside raw audio streaming, OpenAI included direct integration support for standard telecommunications protocols. That allows businesses to connect the model directly to customer service phone numbers and virtual PBX switchboards without routing audio through intermediate conversion servers.
The model also includes tighter steerability for custom voices, tone matching, and strict rule adherence, making it less likely to wander off script or misinterpret caller accents during live interactions.
What This Means for You as a Builder
Building automated phone agents used to mean dealing with frustrating lag and robotic robotic cadence that immediately annoyed callers. With direct API access to speech-to-speech models, building inbound receptionist bots, appointment scheduling lines, and responsive voice agents is now accessible directly with standard web webhooks.