OpenAI launches GPT-Live-1 voice model in API

OpenAI on Thursday released GPT-Live-1 in its API, opening its full-duplex voice model to developers for building voice-enabled applications and business workflows at a cost of $0.05 per minute for the front-end voice layer. The model, which first debuted in ChatGPT in July, can listen and speak simultaneously and delegates deeper reasoning to backend models like GPT-6 Astra or third-party alternatives.openai+1

The release marks a shift from traditional voice-agent architectures that chain together speech-to-text, a language model, and text-to-speech — a pipeline where each handoff adds latency and risks losing conversational rhythm. GPT-Live-1 instead processes incoming and outgoing audio in a single model, responding to interruptions and acknowledgements in real time while routing complex tasks to the backend.openai

Performance Gains Over Previous Models

OpenAI reported that GPT-Live-1 improves Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1, with turn-taking latency dropping from 1.41 seconds to 0.798 seconds. Paired with GPT-6 Astra at medium reasoning effort, the model ranks first on Tau3, a benchmark for voice-agent intelligence on end-to-end tasks, achieving an 86.2% pass rate on spoken customer-service scenarios compared with 45.7% for its predecessor.unite

The API release includes controls for tone, pace, and conversational style through system prompts, along with telephony support for deploying voice agents over phone lines. OpenAI is also expanding its voice library across accents, dialects, and languages.openai

Early Adopters Already in Production

Yelp ↘ 6.49% announced it has integrated GPT-Live-1 into two products: Yelp Host, its voice AI for restaurants, and Hatch, its communication platform for service businesses. Yelp Host has handled more than one million calls since launching in October 2025, and initial production testing with the new model showed improved call-handling rates and fewer call transfers.markets.financialcontent+1

"Callers are also speaking fuller, more natural sentences, which tells us the experience on the other end of the phone feels genuinely different," said Yelp chief technology officer Alex Levy.unite

Language-learning platform Speak reported that GPT-Live-1 cut interruptions during learners' thinking pauses by almost 80% compared with previous turn-based systems. Tony Stoyanov, co-founder and CTO of a healthcare company, said the model simplified his team's codebase by 80%, removing 23,000 lines of code.investing+1

Pricing and Enterprise Access

Developers can pair GPT-Live-1 with the backend model of their choice — OpenAI suggested using Luna for high-volume tasks like scheduling and Astra for complex reasoning — paying only the $0.05-per-minute voice-layer fee on top of standard backend costs. For enterprises, OpenAI pointed to its Presence product, introduced in July, which uses GPT-Live-1 to power deployed voice agents through a managed program rather than self-serve access.unite+1