Does the whisper-1 retirement break SRT and VTT output?

Last updated: October 3, 2026

⏳ 144 days left — the old API shuts down February 26, 2027.

It can. OpenAI removes whisper-1 on February 26, 2027. Its API reference lists json as the only response format for gpt-4o-transcribe and gpt-4o-mini-transcribe, and its guide documents word timestamps only for whisper-1. Test srt, vtt and timestamps on the replacement before you change the model name. If you need the exact whisper-1 request and output, point the OpenAI SDK at a whisper-1-compatible endpoint, or use a URL-to-subtitles tool when your input is a public media file.

Facts on this page were checked against OpenAI's deprecations page, speech-to-text guide and API reference, and the two Actors' READMEs, on October 3, 2026.

What OpenAI's documentation says

Removal date. On August 26, 2026, OpenAI notified developers that whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize will be removed on February 26, 2027. It recommends gpt-transcribe or gpt-live-transcribe (deprecations).
Formats by model. The transcription reference lists json, text, srt, verbose_json, vtt and diarized_json. "For gpt-4o-transcribe and gpt-4o-mini-transcribe, the only supported format is json". gpt-4o-transcribe-diarize supports json, text and diarized_json. The reference does not list subtitle formats for gpt-transcribe.
Word timestamps. "The timestamp_granularities[] parameter is only supported for whisper-1", and it requires response_format=verbose_json (speech-to-text guide).
Translation. The guide uses /v1/audio/translations with whisper-1.

Developers already hit subtitle problems with whisper-1 itself, for example SRT generation and VTT timestamps that do not line up (Stack Overflow). A migration is a good time to test those outputs properly rather than assume them.

What to test before you switch

You use	Check on the replacement
`response_format=srt` or `vtt`	Is the format accepted? Do cue boundaries and timing match your player and caption pipeline?
`verbose_json` with `timestamp_granularities[]=word`	Are word timestamps available at all? If not, what replaces karaoke captions, search-in-audio or editing tools?
`/v1/audio/translations`	Is there an equivalent path to English, or do you need a separate translation step?
Language detection and the `language` field	Same language codes, and the same behavior on mixed-language audio?
Error handling	Do error codes and messages still match what your code parses?

Run a handful of your real files through both before changing production. A sample that matches on clean speech can still differ on noise, music or accents.

Two ways to keep subtitles working

Your input	Option	Output	Price
You send audio files through the OpenAI SDK	whisper-1-compatible endpoint: keep `model="whisper-1"`, change `base_url` and `api_key`	`json`, `text`, `srt`, `vtt`, `verbose_json` with word and segment timestamps; `/v1/audio/translations`	$0.006 per minute, billed per started 15 s
You have a direct public URL to a media file	Media URL transcriber: send the URL, get every format back in one row	`text`, `segments`, `srt`, `vtt` (segment timestamps, not word timestamps)	$0.006/min (`base`) or $0.014/min (`small`)

Both run the open Whisper models in 99 languages, not OpenAI's hosted models. Results are close to whisper-1 on clean speech but not identical:

The compatible endpoint runs Whisper small, which measured about 4% word error on clean English (LibriSpeech). It accepts up to 25 MB and 10 minutes per file.
The URL transcriber accepts up to 1 GB per file, and up to 180 minutes with maxMinutesPerFile. It does not download from YouTube, TikTok or other platform pages.

diff

- client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
+ client = OpenAI(base_url="https://dropin-apis--whisper-compat.apify.actor/v1", api_key=os.environ["APIFY_TOKEN"])
  srt = client.audio.transcriptions.create(model="whisper-1", file=f, response_format="srt")

FAQ

Will my SRT and VTT code keep working after February 26, 2027?

Not with whisper-1 on OpenAI, because the model is removed that day. gpt-4o-transcribe and gpt-4o-mini-transcribe only return json, and are removed the same day. Test subtitle output on gpt-transcribe, or keep the whisper-1 request on a compatible endpoint.

Does gpt-transcribe support word timestamps?

OpenAI's speech-to-text guide documents timestamp_granularities[] only for whisper-1. Check the current guide and test before you rely on word timestamps with another model.

Is a whisper-1-compatible endpoint the same model as whisper-1?

No. It keeps the request and the response format. The transcription comes from open Whisper models, so the words can differ on difficult audio.