OpenAI Includes GPT-Live-1 Model in Developer API
The new AI voice model offers simultaneous listening and speaking capabilities along with continuous streaming and tool integration.
OpenAI has made the GPT-Live-1 model available to developers, enabling the creation of more natural voice experiences.
GPT-Live-1 Made Available via API
OpenAI has released the new GPT-Live-1 model via API for those wanting to develop voice-focused applications and business processes.
Previously announced within ChatGPT, this powerful model has the capacity to both listen and speak at the same time.
Advanced Voice Capabilities and Features
The new version has been launched with a focus on allowing developers to guide and personalize voice experiences.
While tone, speed, and style can be adjusted through system prompts, background noise management is also successfully handled.
Interruption Management and Tool Delegation
Thanks to the architecture that processes incoming and outgoing audio through a single model, interruption management has significantly improved compared to previous systems.
The model can easily delegate deeper reasoning and actions to backend text models and tools such as GPT-6 Astra.
Performance and Cost Details
GPT-Live-1 managed to increase Full Duplex Bench performance by 30 percentage points compared to the GPT-Realtime-2.1 model.
The usage cost in the API for the front-end audio layer has been set at $0.05 per minute.