What happened
OpenAI launched the API for GPT-Live-1, its latest real-time full-duplex voice model, allowing developers to integrate simultaneous listening and speaking capabilities into their applications. The model, which debuted in July within the ChatGPT interface, is now accessible for external development, featuring improved interruption handling and lower latency compared to previous Realtime API versions.
OpenAI has released the API for GPT-Live-1, a real-time full-duplex voice model that allows software to listen and generate speech simultaneously. This capability moves beyond the half-duplex constraints of earlier systems, where users had to wait for the AI to finish speaking before responding. The model was initially available only through the ChatGPT interface in July but is now open to developers via API calls.
The company highlights smooth interruption handling as a key strength, citing early evaluations from a partner named Speak. These evaluations indicated that GPT-Live-1 reduced interruptions by nearly 80 percent compared to previous turn-based systems, giving users more time to think before the AI responds. The model is designed to work alongside backend models like GPT-6 Astra for complex tasks such as information lookup and tool usage.
Performance benchmarks show significant improvements over OpenAI's previous Realtime API. On the Tau3 voice-agent intelligence benchmark, which tests spoken customer-service tasks, GPT-Live-1 scored 86.2 percent, compared to 45.7 percent for GPT-Realtime-2.1. OpenAI also claims a 30 percent performance improvement in Full Duplex Bench metrics and better turn-taking latency.
Yelp is currently using GPT-Live-1 for its Yelp Host service, an AI-driven restaurant reservation system. Yelp's chief product officer stated that the technology makes calls more conversational and responsive, helping restaurants capture revenue opportunities and allowing staff to focus on guests. The specific pricing for the API is $0.05 per minute of usage, excluding the cost of the backend model, with custom voices available through OpenAI's sales team.
Source details: theregister.com ↗
Why it matters
This release lowers the barrier for developers to create natural, fluid voice interactions in business applications. By enabling full-duplex communication, the model addresses the friction of turn-based systems, potentially improving user experience in customer service, education, and other conversational workflows. The documented performance gains over previous models suggest a significant step forward in practical voice AI deployment.
The availability of a full-duplex voice model via API is a significant development for developers building conversational interfaces. It enables more natural human-like interactions, which can be critical in customer-facing applications where responsiveness and fluidity impact user satisfaction.
The reported performance gains, particularly in interruption handling and benchmark scores, suggest that voice AI is becoming more robust and reliable for real-world business use cases. This could accelerate the adoption of voice-first interfaces in sectors like retail, hospitality, and education.
By offering a clear pricing structure and integration with its enterprise platform, OpenAI is positioning this tool as a practical solution for businesses looking to enhance their digital customer experience. The collaboration with Yelp provides a concrete example of how the technology can be applied to solve specific business problems, such as managing phone reservations.
What to watch next
Monitor how early adopters like Yelp implement the technology in production environments and whether the reported performance metrics hold up in independent testing. Watch for broader enterprise adoption and potential updates to pricing or custom voice options as the API matures.
Independent verification of the performance claims, particularly the 80 percent reduction in interruptions and the benchmark scores, will be important to assess the real-world impact of the model.
The extent to which other major companies adopt GPT-Live-1 for their own voice services will indicate the model's market traction and competitive advantage over other voice AI solutions.
Potential updates to the API, such as new features, pricing adjustments, or expanded language support, could further influence its adoption rate among developers and enterprises.