OpenAI Unveils GPT-Realtime Speech API Out of Beta with New Voices and 20% Price Cut
OpenAI released gpt-realtime and moved its Realtime API out of beta: a low-latency, single-model speech-to-speech system that supports images, remote MCPs, SIP phone calling, reusable prompts and async function calls. New voices (Cedar, Marin) and voice updates improve naturalness. Trained for better instruction-following, function-calling and audio reasoning, the Realtime API is production-ready and priced ~20% lower than the previous preview.