OpenAI has introduced a new suite of models—Realtime-2, Realtime-Translate, and Realtime-Whisper—designed to enhance the capabilities of AI voice agents. Unlike previous unified approaches, this system splits voice processing into discrete, specialized models.

By separating these tasks, the architecture significantly reduces the orchestration overhead—the complex, resource-heavy coordination required to link different AI components together. This allows for more efficient and cost-effective deployment of voice agents in enterprise settings. The update incorporates reasoning capabilities described as GPT-5-class, enabling more complex decision-making during real-time interactions.

OpenAI has introduced a new suite of models—Realtime-2, Realtime-Translate, and Realtime-Whisper—designed to enhance the capabilities of AI voice agents. Unlike previous unified approaches, this system splits voice processing into discrete, specialized models.

By separating these tasks, the architecture significantly reduces the orchestration overhead—the complex, resource-heavy coordination required to link different AI components together. This allows for more efficient and cost-effective deployment of voice agents in enterprise settings. The update incorporates reasoning capabilities described as GPT-5-class, enabling more complex decision-making during real-time interactions.