SunsetDiff

whisper-1 and GPT-4o transcribe shutdown on February 26, 2027

The dataset has 5 OpenAI transcription ids with shutdown date 2027-02-26: gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, whisper-1. OpenAI's deprecations page says: "On August 26, 2026, we notified developers using whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize of their deprecation and removal from the API on February 26, 2027." The recommended replacement on each row is gpt-live-transcribe or gpt-transcribe. The same page says: "For information about the recommended replacements, see the transcription guide." The file transcription page says to start with gpt-transcribe for recorded speech. The realtime transcription page says to start with gpt-live-transcribe for live audio. OpenAI's deprecations page also says: "At the time of the shut down, the model or endpoint will no longer be accessible."

Last reviewed . Publisher: SunsetDiff.

Steps

  1. Read the five ids. The table lists whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, and gpt-4o-transcribe-diarize. gpt-4o-mini-transcribe-2025-03-20 is not in this table. Its stored shutdown date is 2027-01-20.
  2. Pick the call, not only the string. A completed file stays on POST /v1/audio/transcriptions with gpt-transcribe. Live audio from a microphone, call, or media stream moves to a realtime transcription session with gpt-live-transcribe. gpt-live-transcribe is not a supported model for the file endpoint.
  3. Test before the date. Call the replacement that matches the audio you send now. Do that before February 26, 2027. Speaker labels are documented on gpt-4o-transcribe-diarize. Word timestamps are documented on whisper-1. The realtime page says gpt-live-transcribe does not return either.

Models shutting down

These rows are the OpenAI transcription ids in the dataset with shutdown date 2027-02-26. The model id links to its SunsetDiff model page. Deprecation date, shutdown date, and replacement are the fields stored on that row. Not published means the field is empty.

OpenAI's deprecations page says: "On August 26, 2026, we notified developers using whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize of their deprecation and removal from the API on February 26, 2027." The section date on that page is August 26, 2026. These rows store 2026-08-26 as announced_date. deprecation_date is empty. The same page says: "When we announce that a model or endpoint is being deprecated, it immediately becomes deprecated. All deprecated models and endpoints will also have a shut down date."

The shutdown table on that page lists whisper-1, gpt-4o-transcribe, and gpt-4o-transcribe-diarize with replacement gpt-live-transcribe or gpt-transcribe. The gpt-4o-mini-transcribe row lists snapshot gpt-4o-mini-transcribe-2025-12-15 in the model column, and the same replacement. That snapshot is its own row here. It is not a model that stays up after February 26, 2027.

gpt-4o-mini-transcribe-2025-03-20 is not in this table. The January 20, 2027 deprecations table names gpt-4o-mini-transcribe-2025-12-15 as its replacement. That snapshot then shuts down on 2027-02-26. The stored row is gpt-4o-mini-transcribe-2025-03-20.

5 OpenAI transcription ids with shutdown date 2027-02-26.
Model idDeprecation dateShutdown dateRecommended replacement
gpt-4o-mini-transcribeNot published2027-02-26gpt-live-transcribe, gpt-transcribe
gpt-4o-mini-transcribe-2025-12-15Not published2027-02-26gpt-live-transcribe, gpt-transcribe
gpt-4o-transcribeNot published2027-02-26gpt-live-transcribe, gpt-transcribe
gpt-4o-transcribe-diarizeNot published2027-02-26gpt-live-transcribe, gpt-transcribe
whisper-1Not published2027-02-26gpt-live-transcribe, gpt-transcribe

Source: OpenAI deprecations , checked November 3, 2026.

What breaks

On February 26, 2027 these ids reach the shutdown date stored in the dataset. The file transcription page says: "whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize, and /v1/audio/translations are deprecated and will shut down on February 26, 2027." OpenAI's deprecations page says: "At the time of the shut down, the model or endpoint will no longer be accessible." The same page uses sunset and shut down for that moment. It does not print an HTTP status for it.

A 400 from a field the replacement rejects is a different failure. This guide does not invent an error body. The file transcription page says: "For gpt-transcribe, languages replaces the singular language field. Don't send both fields." It also says: "Keep each keyword on one line and don't include <, >, a carriage return, or a line feed. The API rejects the entire request when it encounters one of these characters or when prompt exceeds the model's length limit." It does not print the error sentence.

Sources: OpenAI deprecations , OpenAI file transcription , checked November 3, 2026.

Find the model ids in your code

Search the repository for every id in the table. A file transcription helper and a live caption session often live in different files. How to find every model ID in your codebase covers SDK argument names. The scan page builds a search from dataset ids that have a shutdown date.

OpenAI's deprecations page says to call the List models endpoint, GET /v1/models, and read each model's optional shutdown_date field. When present, the field is a YYYY-MM-DD string, or null if no shutdown date has been announced. The same field is on the Retrieve a model endpoint.

You can also read one id from the free SunsetDiff lookup file. whisper-1 is one of these rows. The API page lists the other files. No key and no signup. The alerts page is the free email form. That page says what the form stores and the one welcome email it can send.

ripgrep

rg -n -w -e 'gpt-4o-mini-transcribe' -e 'gpt-4o-mini-transcribe-2025-12-15' -e 'gpt-4o-transcribe' -e 'gpt-4o-transcribe-diarize' -e 'whisper-1'

Source: OpenAI deprecations , checked November 3, 2026.

Recorded files stay on the transcriptions endpoint

The file transcription page says: "Start with gpt-transcribe. This is the recommended model for transcribing recorded speech in its original language. The specialized examples for speaker labels, word timestamps, and subtitle formats below use deprecated models." It also says: "For audio that is still arriving from a microphone, call, or media stream, use Realtime transcription." Send the audio file to /v1/audio/transcriptions with model gpt-transcribe. The gpt-transcribe model page says: "GPT Transcribe is a speech-to-text model for completed audio files, streamed file transcripts, and committed turns in Realtime sessions over WebSocket." That page lists v1/audio/transcriptions and v1/realtime/transcription_sessions as supported. It lists v1/audio/translations, v1/realtime, and v1/audio/speech as not supported. It lists streaming as a supported feature.

The file transcription page says: "Files can be up to 25 MB. Supported input formats are mp3, mp4, mpeg, mpga, m4a, wav, and webm." It says: "For gpt-transcribe, languages replaces the singular language field. Don't send both fields." It also says: "Keep each keyword on one line and don't include <, >, a carriage return, or a line feed. The API rejects the entire request when it encounters one of these characters or when prompt exceeds the model's length limit." Use prompt for unstructured context about the recording, keywords for literal terms you expect to hear, and languages for the expected input languages. The page says keywords are hints, not required output.

For a completed file, streaming stays on that same endpoint. The page says: "File transcription can stream partial text while the model processes a completed recording. This doesn't require a Realtime session." And: "Set stream=true with gpt-transcribe. The Transcriptions API returns transcript events as the model transcribes each part of the recording." It also says: "Existing gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize integrations also support file streaming. whisper-1 doesn't." Live audio is a different call. The page says: "For live audio from a microphone, call, or media stream, use the Realtime transcription guide instead of the file-oriented streaming path above. It covers the current transcription-session flow and the recommended realtime path with gpt-live-transcribe."

The file transcription page says: "Existing gpt-4o-transcribe and gpt-4o-mini-transcribe integrations also support prompting. gpt-4o-transcribe-diarize doesn't support prompts." And: "For whisper-1, prompts have a 224-token limit and provide less control than the recommended transcription model." For languages it says: "Whisper supports 98 languages, but accuracy varies by language. Existing models that accept one language hint use language instead of languages."

File transcription shape from the speech-to-text guide. Model id only.

import fs from "fs";
import OpenAI from "openai";

const openai = new OpenAI();
const transcription = await openai.audio.transcriptions.create({
  file: fs.createReadStream("fixtures/audio.wav"),
  model: "gpt-transcribe",
});

Sources: OpenAI file transcription , gpt-transcribe , checked November 3, 2026.

Live audio uses a transcription session

The realtime transcription page says: "Start with gpt-live-transcribe. Use file transcription if your audio is already recorded, or see the Transcription overview to compare the workflows." The gpt-live-transcribe model page says: "GPT Live Transcribe is a streaming speech-to-text model for applications that need low-latency transcript deltas from live audio." That page lists v1/realtime/transcription_sessions as supported. It lists v1/audio/transcriptions as not supported. It also lists v1/audio/translations, v1/realtime, and v1/audio/speech as not supported, and streaming as a supported feature. Putting gpt-live-transcribe on POST /v1/audio/transcriptions is not the call those pages describe.

The realtime transcription page says: "For server-side WebSocket transcription with a standard API key, connect to wss://api.openai.com/v1/realtime?intent=transcription. Use the authentication header shown in the WebSocket guide, but omit model from the URL. That guide's model-based URL creates a conversation session and doesn't accept the transcription update below." After the connection opens, send session.update with type transcription and set the transcription model to gpt-live-transcribe. The page says: "For browser audio, see WebRTC." It says: "This example uses 24 kHz PCM audio." And: "In gpt-live-transcribe transcription sessions, omit audio.input.turn_detection or set it to null. The model doesn't support server_vad or semantic_vad."

The same page says: "gpt-live-transcribe uses languages instead of the singular language field. Don't send both." It also says: "gpt-live-transcribe doesn't return detected-language predictions." Prompt, keywords, and languages are the context fields on that session. The page says: "Start by setting audio.input.transcription.delay and testing against your real audio." The starting points it lists are below. It says: "The exact delay in milliseconds can vary by model configuration, so benchmark with representative audio instead of assuming a fixed timing per level."

session.update from the realtime transcription guide. turn_detection stays null.

{
  "type": "session.update",
  "session": {
    "type": "transcription",
    "audio": {
      "input": {
        "format": { "type": "audio/pcm", "rate": 24000 },
        "transcription": { "model": "gpt-live-transcribe" },
        "turn_detection": null
      }
    }
  }
}

Sources: OpenAI realtime transcription , gpt-live-transcribe , OpenAI deprecations , checked November 3, 2026.

Diarization, timestamps, and translation

The file transcription page says: "Existing integrations can use the deprecated gpt-4o-transcribe-diarize model to identify who speaks during different parts of a recording until February 26, 2027. This specialized speaker-labeling model isn't the recommended model for ordinary file transcription." It says: "Request the diarized_json response format to receive segments with speaker, start, and end metadata. For audio longer than 30 seconds, set chunking_strategy to "auto" or a voice activity detection configuration." It also says: "You can optionally supply up to four short audio references with known_speaker_names[] and known_speaker_references[] to map segments onto known speakers." Reference clips on that page are between 2 and 10 seconds. The page says: "Speaker labeling is available through /v1/audio/transcriptions. It isn't supported in Realtime transcription sessions." The deprecations page names gpt-live-transcribe or gpt-transcribe for gpt-4o-transcribe-diarize. It does not name a current model that returns speaker labels after February 26, 2027.

The realtime transcription page says: "gpt-live-transcribe doesn't return word-level timestamps, speaker labels, or transcription confidence scores. If your application requires timestamps or speaker labels, use a compatible file transcription model or add an application-level fallback." The file transcription page says: "Existing integrations can use the deprecated whisper-1 model for word or segment timestamps until February 26, 2027. The timestamp_granularities[] parameter returns structured timestamp data for captioning and video editing." And: "The timestamp_granularities[] parameter is only supported for whisper-1." Those pages do not say gpt-transcribe returns word timestamps or speaker labels.

The file transcription page says: "The deprecated /v1/audio/translations endpoint uses whisper-1 to translate a completed audio recording into English. It remains available until February 26, 2027. Unlike transcription, which preserves the recording's original language, this endpoint returns English text only." It says: "Migrate to Realtime translation with gpt-realtime-translate at /v1/realtime/translations. This requires a translation session, streamed audio input, and translation events. Changing only the model ID in a /v1/audio/translations request doesn't migrate it." The deprecations table lists /v1/audio/translations on the same shutdown date, with replacement v1/realtime/translations. That endpoint is not a model row in the table above. The stored model page is gpt-realtime-translate.

Sources: OpenAI file transcription , OpenAI realtime transcription , OpenAI deprecations , checked November 3, 2026.

Price on the pricing page

The pricing page's transcription table says: "Prices per 1M tokens unless noted." The estimated cost column is a per-minute price. On that table, gpt-live-transcribe is $0.017 per minute, gpt-transcribe is $0.0045 per minute, gpt-4o-transcribe is $0.006 per minute, gpt-4o-mini-transcribe is $0.003 per minute, and gpt-4o-transcribe-diarize is $0.006 per minute. A row labeled Whisper is $0.006 per minute. The page does not label that row whisper-1.

Input and output on the same table follow the per-1M-token header. gpt-4o-transcribe and gpt-4o-transcribe-diarize list input $2.50 and output $10.00. gpt-4o-mini-transcribe lists input $1.25 and output $5.00. gpt-transcribe, gpt-live-transcribe, and the Whisper row show a dash for input and output. The gpt-transcribe model page lists transcription audio at $0.0045 per minute. The gpt-live-transcribe model page lists realtime audio at $0.017 per minute.

Sources: OpenAI pricing , gpt-transcribe , gpt-live-transcribe , checked November 3, 2026.

Test the replacement before the cutoff

OpenAI's deprecations page says the notice period is time to evaluate the recommended replacement, test application behavior, and finish the migration before the model is no longer available. For a recorded file, test gpt-transcribe on POST /v1/audio/transcriptions before February 26, 2027. For live audio, test gpt-live-transcribe in a transcription session before February 26, 2027.

The pre-migration test checklist is the longer list: evals, rejected fields, token limits, tools, price, and latency.

Source: OpenAI deprecations , checked November 3, 2026.

Related SunsetDiff pages

All guides

Questions

Which transcription ids shut down on February 26, 2027?

The dataset lists gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, whisper-1 with shutdown date 2027-02-26. gpt-4o-mini-transcribe-2025-03-20 is not in that list. Its shutdown date is 2027-01-20. The deprecations page says: "On August 26, 2026, we notified developers using whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize of their deprecation and removal from the API on February 26, 2027."

What is the recommended replacement?

The deprecations page names gpt-live-transcribe or gpt-transcribe for each id in the table. It says: "For information about the recommended replacements, see the transcription guide." The file transcription page says to start with gpt-transcribe for recorded speech. The realtime transcription page says to start with gpt-live-transcribe for live audio. The recommended replacement is the official replacement stored on each row. gpt-4o-mini-transcribe: gpt-live-transcribe, gpt-transcribe. gpt-4o-mini-transcribe-2025-12-15: gpt-live-transcribe, gpt-transcribe. gpt-4o-transcribe: gpt-live-transcribe, gpt-transcribe. gpt-4o-transcribe-diarize: gpt-live-transcribe, gpt-transcribe. whisper-1: gpt-live-transcribe, gpt-transcribe.

Can I only change the model string on /v1/audio/transcriptions?

For a completed file, the file transcription page sends the file to /v1/audio/transcriptions with gpt-transcribe. The gpt-transcribe model page lists that route as supported. The gpt-live-transcribe model page lists v1/audio/transcriptions as not supported and v1/realtime/transcription_sessions as supported. For gpt-transcribe, the file page says: "For gpt-transcribe, languages replaces the singular language field. Don't send both fields."

Do the replacements return speaker labels or word timestamps?

The realtime transcription page says: "gpt-live-transcribe doesn't return word-level timestamps, speaker labels, or transcription confidence scores. If your application requires timestamps or speaker labels, use a compatible file transcription model or add an application-level fallback." The file transcription page says speaker labeling is not supported in Realtime transcription sessions, and that timestamp_granularities[] is only supported for whisper-1. Those pages do not say gpt-transcribe returns speaker labels or word timestamps. gpt-4o-transcribe-diarize and whisper-1 remain available for those jobs until 2027-02-26.

What happens to /v1/audio/translations?

The file transcription page says: "The deprecated /v1/audio/translations endpoint uses whisper-1 to translate a completed audio recording into English. It remains available until February 26, 2027. Unlike transcription, which preserves the recording's original language, this endpoint returns English text only." It says: "Migrate to Realtime translation with gpt-realtime-translate at /v1/realtime/translations. This requires a translation session, streamed audio input, and translation events. Changing only the model ID in a /v1/audio/translations request doesn't migrate it." The deprecations table lists that endpoint on 2027-02-26, with replacement v1/realtime/translations.

What does OpenAI say happens at shut down?

OpenAI's deprecations page says: "At the time of the shut down, the model or endpoint will no longer be accessible." That page does not print an HTTP status for that moment.

Is a deprecation date stored?

No deprecation_date is stored on these rows. The deprecations section is dated August 26, 2026, and the dataset stores 2026-08-26 as announced_date. The page says: "When we announce that a model or endpoint is being deprecated, it immediately becomes deprecated. All deprecated models and endpoints will also have a shut down date."

More guides

Deprecation alerts

We email you before models shut down. Unsubscribe anytime.

Deprecation alert signup
Providers (optional)