Summary
On September 10, 2026, OpenAI launched GPT-Live-1 in the API, a full-duplex voice model that listens and speaks at the same time and delegates deeper reasoning and actions to paired models and tools. OpenAI priced the voice front end at $0.05 per minute and reported a 30 percentage point gain on Full Duplex Bench over GPT-Realtime-2.1, with large improvements in turn-taking latency.
What changed
OpenAI released GPT-Live-1 in the API, a full-duplex voice model priced at $0.05 per minute that processes audio simultaneously instead of chaining speech-to-text, reasoning, and text-to-speech, and pairs with backend models and tools such as Codex and ChatGPT Work.
Why it matters
Collapsing the voice pipeline into one full-duplex model cuts latency and the failure points that make voice agents feel unnatural, lowering the barrier to phone-grade voice agents for support and transactions. Per-minute pricing gives developers a clear cost model for voice-first agent products.
Evidence excerpt
GPT-Live-1 improves Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1, can listen and speak at the same time, and is available in the API at $0.05 per minute.