Blog · 2026-10-30 · By Dhrumil Barot, Founder

OpenAI Whisper API Pricing: $0.006/min and Its Replacement

The OpenAI Whisper API costs $0.006 per minute ($0.36/hour), but whisper-1 shuts down on February 26, 2027. Every OpenAI transcription rate, the gpt-transcribe migration, and per-hour costs vs Google, Deepgram and AssemblyAI.

TL;DR: The OpenAI Whisper API (whisper-1) costs $0.006 per minute of audio, which is $0.36 per hour. OpenAI deprecated whisper-1 on 2026-08-26 and removes it on February 26, 2027. Its replacement, gpt-transcribe, costs $0.0045 per minute ($0.27 per hour), 25% less, and gpt-4o-mini-transcribe is $0.003 per minute until the same shutdown date.

OpenAI Whisper API pricing is a flat per-minute rate: OpenAI bills whisper-1 at $0.006 for every minute of audio sent to the /v1/audio/transcriptions endpoint, regardless of language or output format. Prices verified 2026-10-30 against OpenAI's API pricing page and the whisper-1 model page. Transcribing 1,000 hours costs $360 on whisper-1, $270 on gpt-transcribe and $180 on gpt-4o-mini-transcribe. The detail most Whisper pricing pages miss: whisper-1, gpt-4o-transcribe and gpt-4o-mini-transcribe are all deprecated, so any cost plan built on them has a hard end date.

How much does the OpenAI Whisper API cost per minute?

The Whisper API costs $0.006 per minute. Here is every OpenAI speech-to-text model on the current rate card:

Per-hour figures are the per-minute price times 60. The two gpt-4o models show a tilde because OpenAI bills them by token, not by minute (next section). Look up any of these models in our /models catalog.

Why are gpt-4o-transcribe prices only estimates?

gpt-4o-transcribe and gpt-4o-mini-transcribe are billed per token, and OpenAI publishes a per-minute estimate next to the token rates. gpt-4o-transcribe costs $2.50 per 1M input tokens and $10.00 per 1M output tokens, which OpenAI estimates at $0.006 per minute. gpt-4o-mini-transcribe costs $1.25 and $5.00 per 1M tokens, estimated at $0.003 per minute. The real bill depends on how many audio tokens your recording produces and how long the transcript is, so dense speech costs more per minute than a call with long silences. A prompt also adds input tokens. whisper-1 and gpt-transcribe are billed purely per minute, so their cost is predictable from audio duration alone.

When does whisper-1 shut down?

whisper-1 shuts down on February 26, 2027. OpenAI notified developers on August 26, 2026 that whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize are deprecated, per the OpenAI deprecations page. The recommended replacement for all four is gpt-transcribe for recorded files or gpt-live-transcribe for streams. The /v1/audio/translations endpoint (Whisper's speech-to-English translation) shuts down on the same date, according to the speech-to-text guide.

Two cost consequences follow:

• Moving whisper-1 to gpt-transcribe cuts the per-minute rate from $0.006 to $0.0045, a 25% saving.
• Teams on gpt-4o-mini-transcribe at about $0.003 per minute see a price increase of 50% when they move to gpt-transcribe. On 10,000 hours a month that is $1,800 becoming $2,700.

How do you switch from whisper-1 to gpt-transcribe?

Switching is a one-parameter change on the same endpoint. The request below is the current form from OpenAI's guide:

curl https://api.openai.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: multipart/form-data" \
  --form file=@/path/to/file/audio.mp3 \
  --form model=gpt-transcribe

In the Python SDK:

from openai import OpenAI

client = OpenAI()
with open("call.mp3", "rb") as audio_file:
    transcript = client.audio.transcriptions.create(
        model="gpt-transcribe",
        file=audio_file,
        prompt="A customer support call about a premium plan and account AC-42.",
    )
print(transcript.text)

gpt-transcribe also accepts keywords (for example "premium plan", "AC-42", "billing") and languages hints (for example "en", "fr") for domain terms and code-switched audio. OpenAI notes that its speaker-label, word-timestamp and subtitle examples still use deprecated models, so check that your response_format and timestamp_granularities needs are supported before you migrate a subtitle pipeline.

What limits affect Whisper API cost?

The main limit is file size. Files sent to the transcription endpoint can be up to 25 MB, in mp3, mp4, mpeg, mpga, m4a, wav or webm. At 128 kbps mp3, 25 MB holds about 27 minutes (26,214,400 bytes x 8 / 128,000 bits per second = 1,638 seconds). Longer recordings must be split or re-encoded at a lower bitrate; splitting does not change the per-minute price. To estimate a file's cost before sending it, read its duration with ffprobe:

secs=$(ffprobe -v error -show_entries format=duration -of csv=p=0 call.mp3)
python3 -c "m=$secs/60; print(f'whisper-1 \${m*0.006:.4f}  gpt-transcribe \${m*0.0045:.4f}')"

Rate limits depend on your usage tier. For gpt-transcribe, the model page lists:

gpt-transcribe is not available on the free tier, so a prototype that ran on whisper-1 without a paid account needs billing set up before migration. The model page also lists Batch (v1/batch) support for gpt-transcribe.

Is the Whisper API cheaper than Google, Deepgram or AssemblyAI?

Yes for Google at list price, no for AssemblyAI. Here are list prices per hour of pre-recorded audio, each from the provider's own pricing page:

Sources: Google Cloud Speech-to-Text pricing, Deepgram pricing, AssemblyAI pricing and Mistral API pricing. AssemblyAI prices per hour; the per-minute column divides by 60.

Google's standard v2 rate falls with volume: $0.01 per minute from 500,000 to 1,000,000 minutes, $0.008 to 2,000,000 and $0.004 above that, and each request is rounded up to the nearest second. Deepgram charges extra for features OpenAI bundles or lacks: speaker diarization adds $0.0020 per minute and Keyterm Prompting $0.0013 per minute. Nova-3 Multilingual is $0.0092 per minute.

What does 10,000 hours of transcription cost per month?

A 10,000-hour month is 600,000 minutes, roughly a mid-size call center recording every call. Multiplying the list prices above:

The Google figure is 500,000 minutes at $0.016 ($8,000) plus 100,000 at $0.01 ($1,000). The table compares price only, not word error rate; run your own audio through the two or three cheapest options before you switch. Our gpt-4o-mini-transcribe vs gpt-4o-transcribe test on Hindi audio shows how much quality can differ by language. Model your own volume in the Vynaris cost calculator.

Is it cheaper to self-host Whisper?

Self-hosting the open-weight Whisper model removes the per-minute fee but adds GPU, ops and idle-time costs. At $0.27 per hour of audio on gpt-transcribe, a self-hosted GPU has to transcribe enough hours to beat that; at low or bursty volume, the API usually wins because you pay nothing while idle. We worked through the same break-even logic for LLMs in self-hosting vs API break-even. Self-hosting also keeps the original Whisper weights available after February 26, 2027, which matters if you depend on Whisper's exact output.

When is the more expensive model worth it?

Pay the $0.017 per minute for gpt-live-transcribe only when someone is waiting on the words in real time, such as live captions or a voice agent. For recordings, the file models cost 74% less ($0.0045 vs $0.017). Pay Deepgram's higher list price when you need its add-ons in one call, like diarization or redaction. Don't route transcription across providers just to save fractions of a cent if your downstream code depends on one provider's timestamp or speaker format; reformatting bugs cost more than the saving.

Where Vynaris fits

Transcription is usually step one. The bigger bill is often the LLM call that summarizes or classifies each transcript. Vynaris is an OpenAI-compatible API that routes each of those LLM requests to the cheapest model that can handle it. See Vynaris pricing and our OpenAI API pricing breakdown for the text-model side of the bill.

FAQ

Is the OpenAI Whisper API free?

No. whisper-1 is billed at $0.006 per minute, and gpt-transcribe is not supported on the free tier at all. The open-source Whisper model is free to run on your own hardware.

How much does Whisper cost per hour?

$0.36 per hour of audio ($0.006 x 60). gpt-transcribe costs $0.27 per hour and gpt-4o-mini-transcribe about $0.18 per hour.

Is whisper-1 being discontinued?

Yes. OpenAI deprecated whisper-1 on August 26, 2026 and removes it from the API on February 26, 2027. Use gpt-transcribe or gpt-live-transcribe instead.

Is gpt-4o-transcribe more expensive than Whisper?

At OpenAI's estimate, no: both are about $0.006 per minute. gpt-4o-transcribe is billed per token ($2.50 in, $10.00 out per 1M), so long, dense transcripts can cost more than the estimate.

Does OpenAI charge for silence in audio files?

whisper-1 and gpt-transcribe are priced per minute of audio, so silence is billed like speech. Trim long silences before upload to cut the bill.

What is the cheapest OpenAI transcription model?

gpt-4o-mini-transcribe at about $0.003 per minute, until its February 26, 2027 shutdown. After that, gpt-transcribe at $0.0045 per minute is OpenAI's cheapest file transcription model.