Imagen

Whisper API — Pricing, Limits and Providers

Last updated Sep 23, 2026 · By ModelsLab Engineering

The ModelsLab Whisper API runs Whisper large-v3 on our own GPUs. POST an audio URL to /api/v6/whisper/transcribe and get the transcript with word-level timestamps. Each request costs $0.0047 from your plan, whatever the length of the audio. For comparison, OpenAI whisper-1 costs $0.006 per minute and Groq whisper-large-v3-turbo $0.04 per hour (checked September 23, 2026).

Get API key See pricing

$0.0047 per request
Whisper large-v3 on our own GPUs
Unlimited on the $149 plan

What the Whisper API Does

Whisper large-v3, Hosted

Whisper is OpenAI’s open-weight speech recognition model. We run Whisper large-v3 on our own GPUs. Running it yourself means a GPU, a queue, a warm deployment and somebody to own all three. This endpoint is the same model with none of that: you POST the URL of an audio file and read the transcript out of the response.

For transcription as a general API, with word timestamps, pricing and a quick start, see our Speech to Text API page.

Because the weights are open, this is not a lock-in decision. The same model you call here is the one you can run on your own hardware later, which is the main reason teams pick Whisper over a proprietary transcription API in the first place.

Billed per Request, Not per Minute

Most transcription APIs meter by the minute or the hour of audio, so a long recording costs proportionally more. ModelsLab bills transcription per request, $0.0047 from your plan, so the length of the file does not change the price of the request. That changes the arithmetic for podcasts, lecture capture, call archives and anything else measured in hours rather than clips. For many short clips, a per-minute vendor such as Groq is cheaper.

Whisper large-v3 over plain HTTP — no SDK to install
Word-level or sentence-level timestamps for subtitles
Multilingual, with an optional language hint
Webhook delivery for long files
$0.0047 per request; no per-request charge on the $149 plan

Whisper API quick start

One POST with an audio URL. Python, JavaScript and cURL.

Python — transcribe a file

Python

1import requests
2
3response = requests.post(
4    "https://modelslab.com/api/v6/whisper/transcribe",
5    json={
6        "key": "YOUR_MODELSLAB_API_KEY",
7        "init_audio": "https://example.com/interview.mp3",
8        "language": "english",
9        "timestamp_level": "word",
10    },
11)
12
13print(response.json())

JavaScript — transcribe with a webhook

JavaScript

1const response = await fetch(
2  'https://modelslab.com/api/v6/whisper/transcribe',
3  {
4    method: 'POST',
5    headers: { 'Content-Type': 'application/json' },
6    body: JSON.stringify({
7      key: 'YOUR_MODELSLAB_API_KEY',
8      init_audio: 'https://example.com/episode-142.mp3',
9      // Long files queue; the webhook receives the finished transcript.
10      webhook: 'https://your-app.example.com/hooks/transcript',
11      track_id: 142,
12    }),
13  },
14);
15
16console.log(await response.json());

cURL — quick test from the terminal

bash

1curl -X POST 'https://modelslab.com/api/v6/whisper/transcribe' \
2  -H 'Content-Type: application/json' \
3  -d '{
4    "key": "YOUR_MODELSLAB_API_KEY",
5    "init_audio": "https://example.com/voicemail.wav",
6    "language": "english"
7  }'

Request and Response

The full contract of the transcription endpoint.

Field	Type	Meaning
init_audio	URL (required)	Audio file to transcribe
language	string (optional)	Source language; defaults to English
timestamp_level	string (optional)	word or sentence timings
webhook	URL (optional)	Where to POST the finished transcript
track_id	string (optional)	Your own numeric reference id, echoed back
output	response	The transcript, with timings when asked

The same handler is registered at /api/v6/voice/speech_to_text; /api/v6/whisper/transcribe is the alias. A v7 route at /api/v7/voice/speech-to-text takes an explicit model_id.

What Transcription Costs

One ModelsLab subscription covers transcription, speech, image and video. A transcript is one request, not a per-minute charge.

Plan	Price / Month	Transcription	Whisper included
Basic	$21	$0.0047 per request	Yes
Standard	$47	$0.0047 per request	Yes
Open Source Unlimited	$149	No per-request charge	Yes

Plan prices from the live plans table, 2026-09-23. A transcription request draws $0.0047 from the plan's dollar allowance. Open Source Unlimited covers the open-weight speech models ModelsLab self-hosts, which includes Whisper large-v3.

Whisper API Providers Compared

Hosted Whisper and its closest rivals, priced per hour of audio. Every price read on the provider's own page on September 23, 2026.

Provider	Model	Listed price	Per hour of audio
Groq	whisper-large-v3-turbo	$0.04 / hour (10 s minimum)	$0.04
Together AI	Whisper Large v3	$0.0015 / minute	$0.09
Groq	whisper-large-v3	$0.111 / hour	$0.111
AssemblyAI	Universal-2	$0.15 / hour	$0.15
OpenAI	gpt-4o-mini-transcribe	~$0.003 / minute	~$0.18
ElevenLabs	Scribe v2	$0.22 / hour	$0.22
Deepgram	Nova-3 pre-recorded (pay as you go)	$0.0043 / minute	$0.258
OpenAI	whisper-1	$0.006 / minute	$0.36
ModelsLab	Whisper large-v3	$0.0047 / request	$0.0047 per request, any length

Per-hour figures are the listed per-minute price times 60. ModelsLab bills per request rather than by audio length, so its per-hour cost falls as files get longer. The row shows the $0.0047 request price on Basic and Standard.

Which Whisper API is cheapest?

Per hour of audio, Groq is the cheapest hosted Whisper we found: $0.04 an hour for whisper-large-v3-turbo, with every request billed for at least 10 seconds. Together AI is next at $0.0015 a minute. OpenAI's whisper-1 is $0.006 a minute, nine times Groq's turbo rate.

ModelsLab is the cheaper choice when files are long: a request draws $0.0047 from your plan, so a one-hour file costs $0.0047 and a two-hour file costs the same. On the $149 Open Source Unlimited plan there is no per-request charge at all. For thousands of 30-second voice notes, a per-minute vendor wins. For live captions or speaker labels, Deepgram and AssemblyAI do what batch Whisper does not.

Whisper API Limits

File size, input type and minimum billing, where the provider publishes them.

Provider	Input	Size limit	Minimum billed
ModelsLab	Audio URL (init_audio)	No published hard limit; webhook for long files	One request
OpenAI	File upload: mp3, mp4, mpeg, mpga, m4a, wav, webm	25 MB	Per minute
Groq	File upload or URL	25 MB (base tier), 100 MB upload (dev tier), 25 MB by URL	10 seconds

OpenAI and Groq limits from their speech-to-text docs, read 2026-09-23. ModelsLab does not publish a hard size or length limit for the shared endpoint; test your longest files before you depend on them, and use the webhook for long recordings.

Whisper price sources

Every competitor price and limit on this page, with the date we read it.

Groq speech-to-text docs — prices, limits and 10 s minimum, read 2026-09-23
OpenAI API pricing — whisper-1 and gpt-4o-mini-transcribe, read 2026-09-23
OpenAI speech-to-text guide — 25 MB limit and formats, read 2026-09-23
Together AI pricing — Whisper Large v3, read 2026-09-23
Deepgram pricing — Nova-3 pre-recorded, read 2026-09-23
AssemblyAI pricing — Universal-2, read 2026-09-23
ElevenLabs API pricing — Scribe v2, read 2026-09-23

Transcribe in 3 Steps

From API key to a transcript.

STEP 01

Step 1: Get Your API Key

Create a ModelsLab account, subscribe to a plan from $21/month, and copy your key from the dashboard.

STEP 02

Step 2: POST the Audio URL

Send init_audio to /api/v6/whisper/transcribe. Pass language when you know it, and timestamp_level when you need word timings.

STEP 03

Step 3: Read the Transcript

Short files return the transcript directly. Long files return an id to fetch, or pass a webhook URL and receive the result when it is ready.

Get API Key

Related speech API guides

Speech to Text API

The same transcription endpoint as a general API, with word timestamps and a quick start.

Best Speech to Text API 2026

How Whisper compares with Deepgram, AssemblyAI, Rev and Google Cloud.

Text to Speech API

The other direction: turn the transcript back into speech in 48 languages.

ElevenLabs API pricing

Per-character TTS pricing, credits and rollover, with the math.

ModelsLab pricing

Audio, image and video plans on one account.

What Whisper Is Not Good At

Whisper is a batch model. It transcribes files well and it does not do live streaming, and it does not label who is speaking. If your product needs real-time captions on a live call, or speaker diarisation out of the box, a specialist streaming vendor will serve you better than this endpoint will — and it is cheaper to know that before you integrate than after.

Why Run Whisper Here

Key advantages that set us apart

Whisper large-v3 on our own GPUs

Billed per request, not per minute of audio

Word-level timestamps for subtitles

Multilingual, with an optional language hint

Webhook delivery for long recordings

Plain HTTP — no SDK, no streaming session

Plans from $21/month; unlimited on the $149 plan

Our Popular Use Cases

What teams build on hosted Whisper:

Transcribe a back catalogue at one API call per episode, then publish the text for search.

Your Data is Secure: GDPR Compliant AI Services

ModelsLab GDPR Compliance Certification Badge

GDPR Compliant

Pricing That's Perfect

Choose plan as per your needs, cancel anytime.

100% refund policy on monthly & yearly plans — cancel anytime

Contact Sales

Best Value

Open Source Unlimited

Mission-Critical

$149/month

100% refund policy · cancel anytime

Unlimited Open Source Models

100% refund policy

24x7 Support

15 parallel generations

Access to all APIs

Unlimited generations on all open-source models

For mission critical workloads

Add Team Members

Priority GPU Clusters

Standard

Production

$47/month

100% refund policy · cancel anytime

Moderate Traffic

100% refund policy

Priority Developer Support

10 concurrent API requests

For Production workloads

API access to all models

Prototype

Basic

Prototype

$21/month

100% refund policy · cancel anytime

Moderate Traffic

100% refund policy

Developer Support via Discord/Email

5 concurrent API requests

API access to all models

Shared GPU

Speech to Text API
Transcription with word-level timestamps.
Text to Speech API
48 languages, billed per second of audio.
Voice Cloning API
Clone a voice from a short sample.
Cheapest text to speech API
TTS cost per minute of audio, compared.
Cheapest AI API
Image, video and speech prices side by side.
Unlimited AI API
Every self-hosted open-source model for $149/mo.

Get Expert Support in Seconds

We're Here to Help.

Want to know more? You can email us anytime at support@modelslab.com

View Docs

POST https://modelslab.com/api/v6/whisper/transcribe with `init_audio` set to the URL of the audio file. `language` is optional and defaults to English, `timestamp_level` takes `word` or `sentence`, and `webhook` receives the result for long files. The same handler is also registered at /api/v6/voice/speech_to_text.

Whisper large-v3, running on ModelsLab's own GPUs.

A transcription request draws $0.0047 from your plan whatever the length of the audio: Basic is $21/month, Standard $47/month, and the $149/month Open Source Unlimited plan has no per-request charge on self-hosted models including Whisper. For comparison, OpenAI whisper-1 costs $0.006 per minute and Groq whisper-large-v3-turbo $0.04 per hour (checked September 23, 2026).

It depends on volume and on whether you already own GPUs. Self-hosting Whisper large-v3 means a GPU, a queue and a deployment to keep warm; on a plan the same work is one HTTP call. Above about 31,700 transcripts a month ($149 ÷ $0.0047), the $149 Open Source Unlimited plan removes the per-request charge, which is usually cheaper than keeping your own GPU warm.

Yes. Set `timestamp_level` to `word` to get per-word timings rather than the sentence-level default, which is what subtitle and caption pipelines need.

Whisper is a multilingual model and the endpoint accepts a `language` hint rather than a fixed list, so pass it when you know the language and leave it out when the input varies. If a specific language matters to your product, test it on your own audio before committing — we would rather you measure it than take a number off a marketing page.

Yes. ModelsLab is a paid service; plans start at $21/month. Creating an account costs nothing, but transcription needs an active plan.

Per hour of audio, Groq: whisper-large-v3-turbo is $0.04 per hour with a 10-second minimum per request. Together AI charges $0.0015 per minute and OpenAI whisper-1 $0.006 per minute (checked September 23, 2026). ModelsLab bills $0.0047 per request from your plan, so it is cheaper for long files: a two-hour recording costs the same as a two-minute one.

OpenAI's transcription API accepts files up to 25 MB, and Groq accepts 25 MB (100 MB by upload on its dev tier), both read September 23, 2026. The ModelsLab endpoint fetches init_audio from a URL instead of taking an upload and does not publish a hard size limit. Use the webhook parameter for long recordings and test your longest files first.

Yes. OpenAI serves whisper-1 at $0.006 per minute, plus gpt-4o-mini-transcribe at about $0.003 per minute (checked September 23, 2026). Whisper's weights are open, so hosts such as ModelsLab, Groq and Together AI serve the same model family.

Send requests.post('https://modelslab.com/api/v6/whisper/transcribe', json={...}) with key and init_audio set to the audio URL, plus language and timestamp_level when you need them. The JSON response carries the transcript in output; long files return an id to fetch, or deliver to your webhook.