OpenAI Launches Realtime API and gpt-realtime Model for Developers

At a glance

OpenAI has made its real-time voice API and voice model gpt-realtime generally available (GA). The model's quality and stability have improved, and developer-friendly features have been added. This post summarizes the changes developers should know about.

Key points

1. 1

   The GA version uses a new interface, and the older beta interface will be discontinued later.

2. 2

   Temperature (a value that controls how random responses are) was removed from the GA version, and the recommended value is 0.8.

3. 3

   A new feature lets the model speak first when a conversation pauses for a long time, such as by asking, "Are you still there?"

4. 4

   The model follows instructions better, so it is a good idea to rewrite prompts (instructions given to the model) for the new model.

Why it matters

It covers API changes and new features that developers building voice AI for phone support or voice assistants need to know right away.