· via dev.to (home feed)
OpenAI to shut down whisper-1 and other transcription models on 26 February 2027
OpenAI will retire whisper-1, three gpt-4o transcription models and the /v1/audio/translations endpoint on 26 February 2027, per a dev.to summary of the API guide. gpt-transcribe is the recommended starting point.

What is shutting down
Developers who transcribe audio through OpenAI's API now have a fixed deadline to work against. According to a post on dev.to that summarises OpenAI's speech-to-text guide, five offerings are deprecated and will stop working on 26 February 2027: the whisper-1 model, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize, and the /v1/audio/translations endpoint used to translate audio into English text.
The post stresses that whisper-1 is only one entry on the list. Applications calling any of the newer gpt-4o transcription models, or the translations endpoint, are just as exposed.
What OpenAI points to instead
The dev.to write-up reports that OpenAI's guide names gpt-transcribe as the recommended starting point for transcribing recorded speech, and notes a 25 MB file-size limit on uploads.
There is a wrinkle: according to the post, the guide's own examples for speaker labels, word timestamps and subtitle formats such as SRT and VTT still run on the deprecated models, kept there so existing integrations keep working until the shutdown. Developers who depend on those outputs may need to verify that the replacement covers them, or plan a different route entirely.
A migration checklist
The post lays out a practical checklist for affected teams:
- Search for every deprecated model name across code, configuration files and no-code automations, not just the obvious API calls.
- Do not move from one deprecated model to another. Switching from whisper-1 to gpt-4o-transcribe achieves nothing if both disappear on the same day.
- Audit the features you rely on. Word timestamps, speaker labels and subtitle output are precisely the examples the guide ties to the deprecated models.
- If you use /v1/audio/translations, plan a separate migration path, since that endpoint is retired alongside the models.
- Test on your own audio. Accents, background noise and domain-specific vocabulary matter more than published benchmarks.
- Re-estimate costs. The post notes that pricing units differ between models, so the same hour of audio can cost a different amount after migration, and points to OpenAI's pricing page for current numbers.
- Set an internal completion date well before 26 February 2027.
A caveat on the reporting
The article's author states it was written with AI assistance and based on OpenAI's documentation read on 18 November 2026, and advises readers to confirm the schedule in the official guide before planning around it, since deprecation dates can shift. That caveat is worth keeping in mind: the date and model list here are as reported by a single secondary source summarising OpenAI's docs, not independently confirmed.
Why it matters
A fixed shutdown date turns a distant API change into a hard break: anything still calling these models on the day will simply fail. The scope is broader than the whisper-1 name suggests, because the retirement covers three gpt-4o-era models and an entire translations endpoint as well. Teams that rely on word-level timestamps, speaker labels or ready-made subtitle files face the most work, since the documentation itself still demonstrates those features on the outgoing models. And with pricing units differing between models, migration becomes a budget exercise as much as an engineering one. The practical takeaway from the dev.to post is to inventory usage now, test the replacement on real audio, and finish well before February 2027.
- #openai
- #whisper
- #api
- #speech-to-text
- #model-deprecation