GPT-Live-Transcribe
GPT-Live-Transcribe produces incremental text from live audio. It is intended for applications that need a transcript before the speaker finishes a full recording.
Hints and latency controls serve live audio
The model accepts context, keyword hints and multiple language hints. The documented rate is $0.017 per audio minute. It uses Realtime transcription sessions and returns text, rather than a spoken response.GPT-Live-Transcribe official model documentation
Separate partial and final text
Practical guidance. For captions, show updates in a way that lets the reader distinguish a provisional phrase from a completed segment. Review specialist vocabulary and language switching with representative recordings. Tune delay only after checking whether earlier text needs frequent correction.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | gpt-live-transcribe |
| Input and output | audio, text in; text outGPT-Live-Transcribe official model documentation |
| API routes | Realtime transcription (v1/realtime/transcription_sessions)GPT-Live-Transcribe official model documentation |
Related models
- GPT-Transcribe
OpenAI · GPT transcription
GPT-Transcribe turns recorded speech into text.
Launch date not confirmedAPI - GPT-Realtime-Whisper
OpenAI · GPT transcription
GPT-Realtime-Whisper is a streaming transcription model for live speech.
Launch date not confirmedAPI