OpenAI released three new audio models in its Realtime API. GPT-Realtime-2 is its first voice model with GPT-5-class reasoning. GPT-Realtime-Translate converts speech from over 70 input languages into 13 output languages. GPT-Realtime-Whisper transcribes speech live as a person talks. GPT-Realtime-2 can speak filler phrases such as "let me check that" while it works. It can call several tools at once and say aloud what it is doing. Developers choose from five reasoning levels, with low as the default. OpenAI says the model retains specialist terms, proper nouns and healthcare vocabulary better. It reports gains over the previous version on two audio reasoning tests. Sessions are monitored by classifiers that can halt harmful conversations.
What changed
Realtime voice sessions were capped at a 32K context window and lacked GPT-5-class reasoning.
What it unlocks
Building voice agents that reason, call several tools at once, and translate or transcribe speech live.
- $32 / 1M audio input tokens
- $0.034 per minute translation
- $0.017 per minute transcription
- 32K to 128K context window
What you need to act on it
- OpenAI API access
- developer integration work
Sources