What it is
GPT Transcribe (gpt-transcribe) is OpenAI's speech-to-text model for finished recordings, streamed file transcripts and committed turns in Realtime transcription sessions. It was released on 28 July 2026. On 26 August 2026 OpenAI told developers that whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize shut down on 26 February 2027, naming gpt-transcribe and the streaming gpt-live-transcribe as replacements.
It takes free-form context, a list of expected keywords and several language hints, and can stream partial text from a finished file. Files are limited to 25 MB in mp3, mp4, mpeg, mpga, m4a, wav or webm.
Our review
GPT Transcribe is a better model than the ones it replaces, and a narrower one. Artificial Analysis measures it at 3.3% AA-WER against 4.0% for GPT-4o Transcribe, at $4.50 per 1,000 minutes instead of $6.00. But OpenAI's guide says word timestamps work only on whisper-1 and speaker labels only on gpt-4o-transcribe-diarize, and both shut down on 26 February 2027.
On accuracy it is mid-table. ElevenLabs Scribe v2 scores 2.2% at $3.67 per 1,000 minutes, Gemini 3.5 Transcribe 2.6% at $5.00, and Microsoft's MAI-Transcribe-2 2.0% at $1.67 while running at 307 times real time, against GPT Transcribe's 55.8. Whisper Large v3 Turbo hosted on Groq costs $0.67 per 1,000 minutes at 4.6%: worse, but a seventh of the price.
The migration bill is not uniform. Teams on whisper-1 or gpt-4o-transcribe save 25%, but gpt-4o-mini-transcribe costs $0.003 a minute, so moving off it raises the bill by half. The keyword list helps with product names and jargon, though OpenAI warns the keywords are hints, not required output, and a malformed list rejects the whole request.
Data handling is a genuine strength. OpenAI lists /v1/audio/transcriptions with no abuse-monitoring retention and as eligible for Zero Data Retention, API audio is not used for training by default, and transcription supports data residency in the US and Europe. For a clinic or a law firm that is worth more than a point of word error rate.
Where it earns its place
A backend engineer maintaining a whisper-1 integration in a note-taking app
Migrate before the February 2027 shutdown
Change the model name and replace the single language field with the languages list. Cost falls from $0.006 to $0.0045 a minute, but any feature built on timestamp_granularities breaks, because only whisper-1 returns word timestamps.
A compliance lead at a 50-person law firm transcribing client calls
Get transcripts without audio lingering on a vendor's servers
The file transcription endpoint keeps no abuse-monitoring logs and qualifies for Zero Data Retention, and API data is not used for training by default. At 20,000 minutes a month the bill is $90; attributing speakers still needs a person or another tool.
An operations manager at a 200-seat contact centre
Transcribe 500,000 call minutes a month for quality scoring
GPT Transcribe costs $2,250 a month at that volume, against about $835 for MAI-Transcribe-2 and $1,835 for Scribe v2 at Artificial Analysis's listed prices. It does not separate agent and customer turns, which most quality scoring needs.
Before you commit
- Setup
- One POST to /v1/audio/transcriptions with model set to gpt-transcribe. Files over 25 MB must be compressed or split, ideally at sentence boundaries, carrying context forward in the prompt.
- Integrations
- File transcription and Realtime transcription sessions only; not available through Chat Completions, Responses, Batch or the translations endpoint. Output is JSON with the text and detected languages.
- Data and privacy
- API audio is not used for training unless you opt in. The /v1/audio/transcriptions endpoint has no abuse-monitoring retention and is eligible for Zero Data Retention; transcription sessions support data residency in the US and Europe.
- Cost at scale
- 100,000 minutes a month costs $450 with no volume or Batch discount. The same volume of live audio on gpt-live-transcribe costs $1,700.
- Limits
- 25 MB per file. 5,000 requests a minute on the Build tier, 10,000 on Launch and 30,000 on Grow. No word timestamps or speaker labels; invalid keywords or an over-long prompt reject the whole request.
What's good
- 3.3% AA-WER, better than GPT-4o Transcribe at 25% lower price
- No abuse-monitoring retention on the transcription endpoint; ZDR eligible
- Keyword, context and multi-language hints
- Works in Realtime transcription sessions
What's not
- No word timestamps or speaker labels
- Less accurate than Scribe v2, Gemini 3.5 Transcribe and MAI-Transcribe-2
- 50% dearer than gpt-4o-mini-transcribe
- 25 MB file limit and no Batch discount
Best for
Teams on the OpenAI API who must leave Whisper or GPT-4o Transcribe before February 2027 and do not need speaker labels or timestamps.
Alternatives
GPT Transcribe: common questions
- Is GPT Transcribe worth it?
- A sensible upgrade for OpenAI shops, though the deprecation it rides on takes away speaker labels and timestamps that this model does not give back. We rate it 3.9 out of 5.
- How much does GPT Transcribe cost?
- $0.0045 per minute of audio ($4.50 per 1,000 minutes); no Batch support. Streaming gpt-live-transcribe is $0.017 per minute.
- What are the downsides of GPT Transcribe?
- No word timestamps or speaker labels; less accurate than Scribe v2, Gemini 3.5 Transcribe and MAI-Transcribe-2; 50% dearer than gpt-4o-mini-transcribe; 25 MB file limit and no Batch discount.
- Who is GPT Transcribe best for?
- Teams on the OpenAI API who must leave Whisper or GPT-4o Transcribe before February 2027 and do not need speaker labels or timestamps.
- How much does GPT Transcribe cost for a team or in production?
- 100,000 minutes a month costs $450 with no volume or Batch discount. The same volume of live audio on gpt-live-transcribe costs $1,700.
- Is GPT Transcribe safe for company data?
- API audio is not used for training unless you opt in. The /v1/audio/transcriptions endpoint has no abuse-monitoring retention and is eligible for Zero Data Retention; transcription sessions support data residency in the US and Europe.
- What are the best GPT Transcribe alternatives?
- The closest alternatives we have reviewed are ElevenLabs Scribe v2, Gemini 3.5 Transcribe and Deepgram Nova-3.
Answers checked Oct 23, 2026.