GPT Realtime Translate, Explained Simply
What GPT Realtime Translate is, what it does well, and what it costs — released 2026-05-07 by OpenAI.
- OpenAI released a new audio model designed specifically to translate spoken language in real-time.
- It processes "streaming" audio, meaning it translates words almost instantly as you talk, without waiting for you to finish.
- It is built for developers who want to create apps for live conversations between people speaking different languages.
Want a heads-up when OpenAI Audio updates?
We email you the moment a new OpenAI Audio version drops — plus plain-English release notes like these. No spam.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
OpenAI just dropped a new audio model made specifically for live, spoken translation. It’s designed for developers who want to build apps where two people speaking totally different languages can chat smoothly, almost as if they had a futuristic robot interpreter sitting between them.
What it does well
This model focuses on "streaming speech translation." In plain English, streaming means the AI doesn't wait for you to finish a whole sentence, pause, and then translate. Instead, it listens to the continuous flow of your voice and starts translating right away. This makes conversations feel much more natural, cutting out those awkward, silent waiting periods that usually happen with translation apps.
How it compares
OpenAI released this alongside two other new audio models: GPT-Realtime-2 and GPT-Realtime-Whisper. While GPT-Realtime-2 is built for general voice agents (like an AI you can just chat with) and GPT-Realtime-Whisper is made for turning speech into text, this model is dialed in specifically for translating speech to speech. It’s a specialized tool rather than a general-purpose voice assistant.
This is our plain-language summary. Read the complete, official notes on the OpenAI Audio official changelog ↗.
Google Calendar