Next Article
OpenAI's new AI model can hold real conversations
OpenAI just introduced gpt-realtime, their most advanced speech-to-speech AI model yet (announced August 28, 2025).
This new tech can sound more natural and expressive, follow complex instructions, and even switch languages or tones in the middle of a sentence—making conversations with AI feel way smoother.
gpt-realtime is now available through the Realtime API
gpt-realtime picks up on things like laughter and gets numbers right in multiple languages (think Spanish, Chinese, Japanese, French). It's built for real-life uses like customer support or personal assistants.
Developers can start using it now through the Realtime API. Plus, there are two fresh voice options—Cedar and Marin—for more variety in how your AI sounds.