Skip to main content

openai/gpt-4o-transcribe/speech-to-text

OpenAI's GPT-4o Transcribe is a speech-to-text model built on GPT-4o. Send a recording up to 25 MB and two hours, add an optional prompt with names and technical terms, and get a transcript in the language spoken.

From $0.0001/s

Workbench

Input

Speech to text
0/2048
Spelling and context in the spoken language at most 2048 characters
1/1
1 / 1 files
Interview recording, up to 25 MB at least 1 characters
Two-letter language of the recording at least 2 characters · at most 2 characters
Sampling temperature between 0 and 1 no lower than 0 · no higher than 1

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result
Model response

Related models and next tasks

Missing the model you need?

RECORDED SPEECH, SPECIALIST CONTEXT

Give technical conversations a written record you can search and check.

Key numbers

Context / structure
prompt
Supply names, instrument labels and specialist terms in prompt so the recognizer has context for the interview.
File limit
25 MB
Interview files must fit both the 25 MB size ceiling and the two-hour duration limit.
Audio limit
2 h
Interview files must fit both the 25 MB size ceiling and the two-hour duration limit.
Minimum billing
1 s
Measured audio seconds

Before you spend

All modes
  • Interview files must fit both the 25 MB size ceiling and the two-hour duration limit.
  • A direct audio file or uploaded recording is needed; an interview's player page is not the media itself.
  • Accents, background noise and overlapping speech can still produce recognition errors in specialist discussions.
  • This standard file endpoint exposes no speaker-label or timestamp controls; prompt supplies context, not a guaranteed correction.
  • OpenAI has deprecated GPT-4o Transcribe and announced its shutdown on 2027-02-26; until then, it remains available with normal billing. For new file-transcription projects, start with GPT Transcribe rather than building around GPT-4o Transcribe.

What it does

A teal tuning fork surrounded by small crystal prisms under warm gold light.

Vocabulary context

Vocabulary before recognition

Supply names, instrument labels and specialist terms in prompt so the recognizer has context for the interview. A short list of real spellings is more useful than directions to rewrite the speaker. Check any rare term against the audio before it enters a technical record.

Vocabulary context

Keep the spoken language

A teal acoustic horn facing a row of blank ceramic panels in a dark gallery.

Set language when you know it; GPT-4o Transcribe writes the recorded speech rather than translating the conversation. The optional language code narrows the recognition context when the recording language is already known. Keep multilingual or uncertain passages available for listening instead of treating the text as a translation.

Two teal glass wave sculptures face a metal inspection lens, linked by a gold path.

Vocabulary context

A transcript for close review

Search the finished text for a claim, then listen to the source before quoting a technical detail or a person's name. Keep a link between the source recording and its transcript in your application. Names and measurements deserve particular attention when the text becomes evidence for a report.

Where it fits

  • Research interviews

    Find a reagent or laboratory mentioned during a recorded research discussion.

  • Engineering handoffs

    Attach a searchable account of a release discussion to the handoff record.

  • Oral history

    Locate a neighborhood name before verifying the speaker's wording.

  • Editorial source work

    Build a working transcript for checking quotations against an interview.

Model reference

GPT-4o Transcribe API reference

GPT-4o Transcribe API specifications

Input
GPT-4o Transcribe · Audio URL · MP3 / MP4 / MPEG / MPGA / M4A / WAV / WEBM
Language
Optional ISO 639-1
Temperature
0–1; default 0
Billing
Measured audio seconds
Delivery
Asynchronous task

How to call the GPT-4o Transcribe API

  1. Choose the passage

    Start with a representative interview excerpt and keep the original recording available.

  2. Add real context

    Provide relevant spellings in the recording's language and select its language code if known.

  3. Review the completed task

    Collect the asynchronous result and check names, numbers and uncertain phrases by listening.

The short version

Provide relevant spellings in the recording's language and select its language code if known. Collect the asynchronous result and check names, numbers and uncertain phrases by listening.

GPT-4o Transcribe API questions

All GPT models18