OpenAI introduced three audio models for its developer platform on Thursday, aiming to make voice-based software agents more conversational and capable of completing tasks in real time.
The launch of the application programming interface (API) moves the ChatGPT-maker beyond transcription and chat towards agents that can listen, translate and act during live conversations.
The new models are GPT-Realtime-2, GPT-Realtime-Translate and GPT-Realtime-Whisper. OpenAI said they are available to test in its developer playground.
GPT-Realtime-2 is designed to manage harder requests, call tools, handle interruptions and maintain context across longer voice sessions.
The second model supports translation from more than 70 languages ​​into 13 output languages, targeting customer support, education and other settings.
GPT-Realtime-Whisper provides live speech-to-text, allowing captions, meeting notes and workflow updates to be generated as a speaker talks.
Customers testing the models include online real estate marketplace Zillow, online travel agency Priceline and European telecommunications firm Deutsche Telekom.
Pricing for GPT-Realtime-2 starts at $32 per million audio input tokens, GPT-Realtime-Translate costs $0.034 per minute and GPT-Realtime-Whisper $0.017 per minute.