GPT-Live-1 Brings Full-Duplex Voice Agents to the OpenAI API
Priced at $0.05 per minute for the voice front end, GPT-Live-1 listens and speaks at once and can hand hard work to paired reasoning models.
1 min read
Also on 10 September 2026, OpenAI launched GPT-Live-1 in the API — the full-duplex voice model first shown in ChatGPT, now available for apps and workflows.
Why full duplex matters
Older voice stacks waited for silence before answering. That feels polite in demos and clumsy in real conversation. GPT-Live-1 is built to listen and speak concurrently, with stronger interruption handling. OpenAI cites early evaluations with language-learning company Speak suggesting nearly 80% fewer interruptions versus prior turn-based systems, and large gains on internal full-duplex benchmarks versus GPT-Realtime-2.1.
Pricing for the voice front end is listed at $0.05 per minute, with deeper reasoning delegated to paired backend models and agent harnesses.
The CTRL product lens
Voice is becoming a primary interface for support, tutoring, and on-device assistants. The interesting question is not “can it talk?” but “can it stay useful when users talk over it, change topics, or ask for actions?” Pairing a live voice layer with an agents harness is OpenAI’s answer to that stack.
Expect a wave of demos — and a shorter wave of products that survive latency, cost, and privacy reviews.
Comments
Loading comments…