OpenAI Launches Real-Time API for Faster, More Natural Conversational AI
OpenAI has launched its real-time API, bringing significant improvements to conversational AI. The 'gpt-realtime' model offers faster, more natural interactions and supports image input and tool calling.
The new model processes speech directly, eliminating the need for text intermediaries. It also recognizes non-verbal cues like laughter and switches languages mid-sentence. OpenAI has introduced two new voices, Cedar and Marin, and enhanced existing ones. The 'gpt-realtime' model responds faster and follows complex instructions better, as demonstrated in benchmarks.
To manage costs during long sessions, OpenAI has introduced new control functions. The model can now call tools asynchronously, ensuring seamless and reliable conversations. Prices for the real-time API are 20% lower than the previous version.
The official release of the real-time API on October 1, 2025, marked the start of the Secure Release Order process with shipowners. Actual usage began between October 1 and November 3, 2025. With its improved performance, natural interactions, and cost management features, the 'gpt-realtime' model is set to enhance productivity and user experience.
