Overview
GPT-Realtime-Whisper is a streaming speech-to-text model for low-latency, live transcription. OpenAI announced it on May 7, 2026, and made it available through the Realtime API. (openai.com)
Capabilities and uses
- Transcribes speech while a person is speaking and streams incremental transcript text.
- A configurable delay lets developers trade latency for transcription accuracy.
- Intended uses include live captions for meetings, classrooms, broadcasts and events, as well as meeting notes and transcription for voice-agent workflows. Summaries and downstream actions require an accompanying application or other components. (openai.com)
Limitations
- Audio is input-only; the model does not generate audio. Images and video are not supported.
- The documented context window is 16,000 tokens, with a maximum of 2,000 output tokens. Function calling, structured outputs and fine-tuning are not supported.
- In generally available Realtime sessions, the transcription prompt parameter is not supported for this model. (developers.openai.com)
Pricing
OpenAI's API price for live transcription is USD $0.017 per minute of audio. Billing is based on audio duration rather than text tokens; the official pricing table does not list separate input-token or output-token rates for this model. Pricing checked November 16, 2026. (developers.openai.com)