OpenAI updates its voice models with GPT-5-level reasoning in API

The new models out today in the API are GPT-Realtime-2 that can handle requests and do stuff for you, while being a natural conversation partner.

Realtime-Translate can handle live translations from 70+ input languages into 13 output languages and keep up with the speaker.

While Realtime-Whisper is a transcriber that turns speech into text as it happens.

Sam Altman notes that it’s mostly young people that prefer voice interactions with ChatGPT, while older people like to type.

Realtime-2 is priced at a whopping $32 for 1 million input tokens and $64 for 1M outputs. The other models are priced cheaply.

The models are only available in the API so far, so no general access within ChatGPT. They can, however, be added to apps in Codex.

Read more: OpenAI’s announcement, on X.com, 9to5Mac, and TechCrunch.