OpenAI Releases GPT-Live-1 Full-Duplex Voice Model to API
The API release gives third-party developers access to a full-duplex voice layer that handles natural interruptions and can delegate to other models, enabling conversational voice agents without building separate speech pipelines. Yelp's deployment shows the model being used in production for real customer-facing calls.
OpenAI made GPT-Live-1 available in its API on September 10, 2026, priced at $0.05 per minute for the front-end voice layer. The model handles listening and speaking in a single pass, eliminating the latency and handoff issues of chained speech-to-text, language model, and text-to-speech architectures. Developers can delegate deeper reasoning and tool calls to backend text models such as GPT-6 Astra or third-party models, and can integrate the model with infrastructure of their choosing. In evaluations with language-learning company Speak, the model cut interruptions by almost 80% compared to previous turn-based systems. On the Full Duplex Bench it improved by 30 percentage points over GPT-Realtime-2.1, and ranked first on the Tau3 voice-agent benchmark when paired with GPT-6 Astra. Yelp is using GPT-Live-1 to power restaurant reservation phone calls, where callers can interrupt, add details, or change direction mid-conversation. The release is also connected to the 1-800-ChatGPT phone service. OpenAI noted it is expanding available voice options across accents, dialects, and languages.
The 1-800-ChatGPT phone service surfaced September 13 as a deployment context for GPT-Live-1, the full-duplex voice model OpenAI released to its API on September 10 at $0.05 per minute.
In evaluations with language-learning company Speak, the model cut interruptions by almost 80% compared to previous turn-based systems, and it ranked first on the Tau3 voice-agent benchmark when paired with GPT-6 Astra.
Yelp is using GPT-Live-1, OpenAI's full-duplex voice model that reached its API on September 10, for restaurant reservation phone calls, where callers can interrupt, add details, or change direction mid-conversation.
OpenAI also reported the model improved Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1 and expanded voice options across accents, dialects, and languages.
OpenAI released GPT-Live-1 to its API on September 10 at $0.05 per minute for the front-end voice layer, making available a full-duplex model that handles listening and speaking simultaneously in a single model rather than chaining separate speech-to-text, language, and text-to-speech components.
Developers can pair it with backend models including GPT-6 Astra or third-party options for reasoning and tool calls. Testing with language-learning company Speak showed it "cutting interruptions by almost 80% versus previous turn-based systems," and the model ranks first on the Tau3 voice-agent benchmark when paired with GPT-6 Astra.
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free