Whisper API
OpenAI's open speech-to-text model; the hosted API retires in 2027
Overview
Whisper is OpenAI's multilingual speech-recognition model, released in 2022 as open-source weights under MIT and offered through the API as whisper-1. The API model costs $0.006 a minute and is scheduled for removal on 26 February 2027, with gpt-transcribe as OpenAI's recommended replacement; the open model can still be run locally for free.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Model sizes from tiny to large-v3, plus turbo
- Transcription, translation into English and language identification
- Runs offline on your own hardware
- Transcriptions and translations endpoints
- Word and segment timestamps
- Files up to 25 MB
- File and streamed transcription
- Prompt, keyword and multiple-language hints
API pricing (pay as you go)
| Model | Price |
|---|---|
| Whisper Listed as 'Whisper' (API id whisper-1 not named on the page). | $0.006 |
| gpt-transcribe | $0.0045 |
| gpt-4o-transcribe Token rates: $2.50 in / $10.00 out per 1M. | $0.006 |
| gpt-4o-mini-transcribe Token rates: $1.25 in / $5.00 out per 1M. | $0.003 |
| gpt-realtime-whisper Live transcription. | $0.017 |
| gpt-live-transcribe Live transcription. | $0.017 |
Use cases
What you can produce with Whisper API
- Transcripts of recorded speech in dozens of languages, locally or through the API
- Non-English speech translated into English text in one pass (multilingual models, not turbo)
- Offline transcription of sensitive audio with whisper.cpp or faster-whisper and no per-minute cost
- Word- and segment-level timestamps for captions and subtitles via whisper-1
- A speech-to-text layer to build into meeting-notes, captioning or voice apps
- A migration path to gpt-transcribe before the whisper-1 API shuts down in February 2027
The open-source Whisper model is still one of the most useful pieces of speech technology available: free, MIT-licensed, multilingual and runnable offline through projects such as whisper.cpp and faster-whisper. It does hallucinate, particularly around silences, and it has no speaker labels, so high-stakes transcripts need a human check.
The hosted whisper-1 API is a different story. OpenAI will remove it on 26 February 2027 and recommends gpt-transcribe instead, so new API projects should not start on it, and existing ones that use it for timestamps, subtitles or translation should plan their migration now.
What we found
Whisper is the speech-recognition model OpenAI released in September 2022, trained on a large and varied audio dataset to transcribe speech in many languages, translate it into English and identify the language spoken. It exists in two forms that are now heading in different directions. The open-source model, under MIT on GitHub, comes in sizes from tiny to large-v3 plus a faster turbo variant, and OpenAI was still merging changes in August 2026. The hosted form is the whisper-1 API model, priced at $0.006 a minute.
That API model is being retired. On 26 August 2026 OpenAI told developers that whisper-1, together with gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize, will be removed from the API on 26 February 2027, and named gpt-transcribe, at $0.0045 a minute, or the live gpt-live-transcribe, at $0.017, as replacements. OpenAI's own guide still sends developers to whisper-1 for word and segment timestamps, subtitle formats and translation into English, so anyone relying on those features should confirm what replaces them before the cut-off. Earlier versions of this page said no deprecation had been announced; that is no longer true.
The open model is unaffected, and it remains the reason Whisper matters. Projects such as whisper.cpp and faster-whisper run it on ordinary hardware, offline and with no per-minute fee, which is why so many transcription products are built on top of it. It has no built-in speaker labels. OpenAI's own documentation says accuracy varies widely by language, and the turbo model is not trained for translation.
The known weakness is fabrication. The Careless Whisper paper by Koenecke and colleagues, presented at ACM FAccT 2024, found that roughly 1% of transcriptions contained whole hallucinated phrases, more often around pauses, and an Associated Press investigation in October 2024 reported invented text turning up in medical transcription. High-stakes audio needs human review.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Whisper API's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Whisper API unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Whisper API directly →
Spotted something out of date? Suggest an update →
Alternatives to Whisper API
More in Video & Audio
