Whisper API pricing: cost per minute, the shutdown date and alternatives
OpenAI's hosted Whisper costs $0.006 per minute, but it is being retired: OpenAI says whisper-1 shuts down on February 26, 2027. This page gives the price, the replacement, and the open-source option. Compare all providers in speech-to-text API pricing compared or price a job with the speech-to-text cost calculator; the topic hub is AI voice, speech and music.
Prices last verified against official provider pages on . How we verify prices. Written by the Chelure Enterprises Editorial Team.
OpenAI transcription prices
| Model | Price (OpenAI pricing page) | Per hour of audio | Status |
|---|---|---|---|
| Whisper (whisper-1) | $0.006 per minute | $0.36 | Deprecated; shuts down February 26, 2027 |
| gpt-transcribe | $0.0045 per minute | $0.27 | Recommended by OpenAI |
| gpt-4o-transcribe | $2.50 input / $10.00 output per 1M tokens; about $0.006 per minute | about $0.36 | Deprecated; shuts down February 26, 2027 |
| gpt-4o-mini-transcribe | $1.25 input / $5.00 output per 1M tokens; about $0.003 per minute | about $0.18 | Deprecated; shuts down February 26, 2027 |
| gpt-live-transcribe / gpt-realtime-whisper | $0.017 per minute | $1.02 | Live transcription |
Per-hour figures are our arithmetic (per-minute price times 60). OpenAI's file transcription guide tells developers to start with gpt-transcribe for recorded speech in its original language, and to use the Realtime API for audio that is still arriving.
Whisper the open-source model
Whisper is also a general-purpose speech recognition model OpenAI released openly: the repository describes it as handling speech recognition, translation and language identification, under the MIT licence for both code and weights. Running it yourself removes the per-minute fee but you pay for the hardware.
| Size | Parameters | Variants | Required VRAM | Relative speed |
|---|---|---|---|---|
| tiny | 39M | tiny.en / tiny | ~1 GB | ~10x |
| base | 74M | base.en / base | ~1 GB | ~7x |
| small | 244M | small.en / small | ~2 GB | ~4x |
| medium | 769M | medium.en / medium | ~5 GB | ~2x |
| large | 1550M | multilingual only | ~10 GB | 1x |
| turbo | 809M | multilingual only | ~6 GB | ~8x |
Relative speed is measured against large on English speech on an A100, so actual speed varies with language, speaking rate and hardware. The research paper is arXiv:2212.04356.
What to do before February 2027
- Move batch transcription to
gpt-transcribe, which is cheaper at $0.0045 per minute. - If you used
/v1/audio/translations, plan the move to Realtime translation; it is a different workflow, not a model-name swap. - Files are limited to 25 MB each, so long recordings need splitting.
- Compare other providers: Mistral, ElevenLabs, xAI and Google quote lower per-hour rates for batch work.
Frequently asked questions
How much does the Whisper API cost?
OpenAI lists Whisper at $0.006 per minute, which is $0.36 per hour. OpenAI's recommended replacement, gpt-transcribe, is $0.0045 per minute ($0.27 per hour).
Is the Whisper API being shut down?
Yes. OpenAI's transcription guide 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.
Is Whisper free?
The open-source model is: the Whisper repository is released under the MIT licence, covering code and model weights. You pay for the hardware to run it instead of a per-minute fee.
What is the file size limit?
OpenAI's file transcription guide says files can be up to 25 MB, in formats including flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav and webm.
Which Whisper model size should I run myself?
The repository lists tiny (39M parameters, about 1 GB VRAM) through large (1550M, about 10 GB) and turbo (809M, about 6 GB, about 8 times faster than large). Pick the smallest that meets your accuracy needs.
What replaces Whisper translation to English?
OpenAI says the deprecated translations endpoint should migrate to Realtime translation with gpt-realtime-translate, which needs a translation session and streamed audio; changing only the model ID does not migrate it.
Sources
- OpenAI API pricing
- OpenAI: file transcription guide (deprecations, 25 MB limit)
- OpenAI Whisper on GitHub (MIT licence, model sizes)
- Radford et al.: Robust Speech Recognition via Large-Scale Weak Supervision (arXiv:2212.04356)
Primary sources opened on November 13, 2026. Prices come from our verified price table (see the methodology).