Reading up on Realtime API

OpenAI rearchitected its WebRTC stack into a split relay-plus-transceiver architecture to reduce latency for voice AI at global scale.

OpenAI released gpt-realtime, a speech-to-speech model in the API, and published a prompting guide detailing techniques for instruction following, voice quality, and tool calling.

Prompt caching reduces latency by up to 80% and costs by up to 90% by reusing computed key/value tensors for repeated prompt prefixes in OpenAI's API.