Home
OpenAI Adds GPT-Live-1 to API With Full-Duplex Voice and Interruption Support for Customer Service Calls
OpenAI has integrated GPT-Live-1 into its API, empowering developers to create applications and business workflows that enable real-time voice interactions. This model facilitates full-duplex conversations, allowing it to listen and speak simultaneously while managing interruptions, pauses, and background noise effectively.
GPT-Live-1 is also optimized for phone-based scenarios, such as restaurant reservations and customer support, enabling full-duplex voice agents. OpenAI notes that developers can integrate it with tools like Codex or leverage OpenAI Presence to build voice applications that query enterprise systems, execute authorized actions, and seamlessly transfer calls to human agents when needed.

Unified speech understanding and output for reduced latency
GPT-Live-1 combines speech recognition and synthesis within a single model, eliminating the need for multiple connections typical of the traditional "speech-to-text - LLM - text-to-speech" pipeline, thereby significantly lowering latency. The model also handles complex reasoning and tool calls via backend text models. Developers can customize tone, speed, and conversation style through system prompts, while configuring backend models, tools, and agent frameworks. OpenAI confirms that GPT-Live-1 supports native speech recognition transcription and response text, featuring capabilities like alphanumeric understanding, keyword bias, and turn detection.
Early assessments indicate that using GPT-Live-1 reduced mis-interruptions during learners' thinking pauses on the language learning platform Speak by nearly 80% compared to the previous turn-based system. Tony Stoyanov, co-founder and CTO of a medical service platform, reported an 80% reduction in code volume, removing approximately 23,000 lines of code. OpenAI's evaluations demonstrate that GPT-Live-1 outperforms GPT-Realtime-2.1 by 30 percentage points in the Full Duplex Bench test and ranks first in the Tau3 end-to-end voice agent task test when paired with GPT-6 Astra at medium reasoning intensity.
Front-end speech layer priced at $0.05 per minute with 12 new voices
Regarding pricing, the front-end speech layer for GPT-Live-1 is charged at $0.05 per minute, equating to roughly $3 (approximately 20.2 RMB) per hour. Costs for backend models and agent tools are billed separately. OpenAI has also expanded real-time voice options by adding 12 new voices: Quartz, Ripple, Vesper, Willow, Stone, Gleam, Meridian, Bossa, Tempo, Beacon, Delta, and Cinder. The company plans to introduce additional voices and language support in the coming months.
Related article
Apple Unveils First Public Beta of iOS 27, Marking Siri’s Major AI Overhaul
Apple has officially released the first public beta version of iOS 27 for users today. This update does not require a developer account. Users simply need to register or log in on the Apple Beta Software Program website and select the corresponding v
Cloudflare Unveils Open Source AI Agent Workspace, Empowering Employees to Build Custom Workflows Without Code
Cloudflare has officially released Cloudflare OS, an open-source AI agent workspace that operates directly in the browser. This platform integrates a wide array of customizable, shareable micro-applications. Far from being a mere wrapper around gener
Musk’s AI Credit Card Checkout Sparks Backlash After OpenAI Exec’s Mockery
Elon Musk recently unveiled Grok’s new shopping capability on social media, explaining that users can snap a photo of their credit card and send it to the chat interface. The AI will then scour the web for the best prices and finalize the purchase on
Related Special Topic Recommendations
Comments (0)
0/500
OpenAI has integrated GPT-Live-1 into its API, empowering developers to create applications and business workflows that enable real-time voice interactions. This model facilitates full-duplex conversations, allowing it to listen and speak simultaneously while managing interruptions, pauses, and background noise effectively.
GPT-Live-1 is also optimized for phone-based scenarios, such as restaurant reservations and customer support, enabling full-duplex voice agents. OpenAI notes that developers can integrate it with tools like Codex or leverage OpenAI Presence to build voice applications that query enterprise systems, execute authorized actions, and seamlessly transfer calls to human agents when needed.

Unified speech understanding and output for reduced latency
GPT-Live-1 combines speech recognition and synthesis within a single model, eliminating the need for multiple connections typical of the traditional "speech-to-text - LLM - text-to-speech" pipeline, thereby significantly lowering latency. The model also handles complex reasoning and tool calls via backend text models. Developers can customize tone, speed, and conversation style through system prompts, while configuring backend models, tools, and agent frameworks. OpenAI confirms that GPT-Live-1 supports native speech recognition transcription and response text, featuring capabilities like alphanumeric understanding, keyword bias, and turn detection.
Early assessments indicate that using GPT-Live-1 reduced mis-interruptions during learners' thinking pauses on the language learning platform Speak by nearly 80% compared to the previous turn-based system. Tony Stoyanov, co-founder and CTO of a medical service platform, reported an 80% reduction in code volume, removing approximately 23,000 lines of code. OpenAI's evaluations demonstrate that GPT-Live-1 outperforms GPT-Realtime-2.1 by 30 percentage points in the Full Duplex Bench test and ranks first in the Tau3 end-to-end voice agent task test when paired with GPT-6 Astra at medium reasoning intensity.
Front-end speech layer priced at $0.05 per minute with 12 new voices
Regarding pricing, the front-end speech layer for GPT-Live-1 is charged at $0.05 per minute, equating to roughly $3 (approximately 20.2 RMB) per hour. Costs for backend models and agent tools are billed separately. OpenAI has also expanded real-time voice options by adding 12 new voices: Quartz, Ripple, Vesper, Willow, Stone, Gleam, Meridian, Bossa, Tempo, Beacon, Delta, and Cinder. The company plans to introduce additional voices and language support in the coming months.
Apple Unveils First Public Beta of iOS 27, Marking Siri’s Major AI Overhaul
Apple has officially released the first public beta version of iOS 27 for users today. This update does not require a developer account. Users simply need to register or log in on the Apple Beta Software Program website and select the corresponding v
Cloudflare Unveils Open Source AI Agent Workspace, Empowering Employees to Build Custom Workflows Without Code
Cloudflare has officially released Cloudflare OS, an open-source AI agent workspace that operates directly in the browser. This platform integrates a wide array of customizable, shareable micro-applications. Far from being a mere wrapper around gener
Musk’s AI Credit Card Checkout Sparks Backlash After OpenAI Exec’s Mockery
Elon Musk recently unveiled Grok’s new shopping capability on social media, explaining that users can snap a photo of their credit card and send it to the chat interface. The AI will then scour the web for the best prices and finalize the purchase on











