OpenAI says its voice system sends audio through a dedicated fast path, while deeper reasoning and tool useA model's ability to call external functions — run code, search the web, edit files — instead of only generating text.Full definition → run asynchronously, meaning in the background rather than inside the audio stream. The design is meant to keep a spoken conversation moving while slower work happens. The company's post links to a longer engineering write-up of the system.
OpenAI also reports cutting voice-session startup from six network round trips to one. A round trip is one request and reply between the app and the server, so fewer of them means less waiting before a session can begin. The post does not say how much total latency fell or how voice quality changed.
Terms in this piece · Glossary
tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
