GPT-Transcribe
OpenAI · Audio model
GPT-Transcribe is OpenAI's speech-to-text model released in late July 2026 for transcribing completed audio files and batch jobs. It costs $0.0045 per minute and accepts hints such as keywords and expected languages. A streaming sibling, GPT-Live-Transcribe, targets live transcription.
Current
ReleasedJul 28, 2026
Context window—
Input → outputaudio, text → text
API price$0.0045 per minute (transcription audio duration)
Consumer access—
Best for
Accurate, cheap transcription of recorded audio files and batch workloads.
Overview
GPT-Transcribe is OpenAI's speech-to-text model released in late July 2026 for transcribing completed audio files and batch jobs. It costs $0.0045 per minute and accepts hints such as keywords and expected languages. A streaming sibling, GPT-Live-Transcribe, targets live transcription.
Strengths and limits
- Cheaper per minute than whisper-1 and gpt-4o-transcribe ($0.006/min)
- Accepts context, keyword hints and language hints
- Works for files, streamed file transcripts and committed Realtime turns
- For true low-latency live captions OpenAI offers a separate model (GPT-Live-Transcribe)
- Speech-to-text only; no speaker output