GPT-Realtime-Whisper
GPT-Realtime-Whisper is a streaming transcription model for live speech. It lets an application balance transcript delay against recognition quality.
Streaming Whisper has its own API contract
This release uses the Realtime transcription endpoint. It accepts audio and text context and returns transcript text. The listed price is $0.017 per minute, with a 16,000-token context and a 2,000-token output limit.GPT-Realtime-Whisper official model documentation
Measure delay against correction frequency
Practical guidance. For live notes, compare the first visible phrase with the final transcript. Track errors in names and numbers as well as total word accuracy. Decide whether the product benefits more from an early partial result or a later stable segment.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | gpt-realtime-whisper |
| Context window | 16,000 tokens Published model context; input and output share this budget GPT-Realtime-Whisper official model documentation |
| Maximum output | 2,000 tokens Published maximum output; bounded by the remaining context GPT-Realtime-Whisper official model documentation |
| Input and output | audio, text in; text outGPT-Realtime-Whisper official model documentation |
| API routes | Realtime transcription (v1/realtime/transcription_sessions)GPT-Realtime-Whisper official model documentation |
Related models
- GPT-Transcribe
OpenAI · GPT transcription
GPT-Transcribe turns recorded speech into text.
Launch date not confirmedAPI - GPT-Live-Transcribe
OpenAI · GPT transcription
GPT-Live-Transcribe produces incremental text from live audio.
Launch date not confirmedAPI