Build Hour: GPT-Realtime-2
- Source
- youtube.com
- Author
- OpenAI
- Date
Context
In a developer session, OpenAI staff describe three new audio models, including GPT-Realtime-2, which they call their most intelligent voice model. They say it brings GPT-5-class reasoning to voice, accepts a four times larger (128k), can call tools in parallel instead of one after another, and can say a short preamble before it reasons. The other two are a live translation model and a transcription model.
The demo is a shopping assistant that does more than talk. Given roughly 15 or 20 tools, a number the presenter says OpenAI would not have recommended for its previous real-time models, it searched for tents, read low-star reviews, checked a weather forecast, added items to a cart and opened product pages by operating the site itself. The presenters describe this as a step beyond speech in, single action out. The gains they mention were not shown in the part read, so the performance claims are OpenAI's own.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…






