Provider

Free Whisper API

Whisper is the open speech recognition model from OpenAI. You can run it free on Groq, Cloudflare Workers AI and OVH AI Endpoints, each with its own limits. FreeLLMAPI puts these Whisper hosts behind one OpenAI-compatible endpoint. Your code calls /v1/audio/transcriptions, and the router moves to the next host when one hits its limit.

Where Whisper is free

The table lists all free transcription models in the FreeLLMAPI catalog, from the live catalog. The Whisper rows are the ones with whisper in the model id.

ModelProviderMax uploadFree limit
Whisper Large v3 Turbo (Groq)
whisper-large-v3-turbo
Groq25 MBfree tier - 25 MB uploads
Whisper Large v3 (Groq)
whisper-large-v3
Groq25 MBfree tier - 25 MB uploads
Whisper Large v3 Turbo (Cloudflare Workers AI)
@cf/openai/whisper-large-v3-turbo
Cloudflare Workers AI-Workers AI free allocation
Whisper (Cloudflare Workers AI)
@cf/openai/whisper
Cloudflare Workers AI-Workers AI free allocation
Deepgram Nova-3 (Cloudflare Workers AI)
@cf/deepgram/nova-3
Cloudflare Workers AI-Workers AI free allocation
Whisper Tiny English (Cloudflare Workers AI)
@cf/openai/whisper-tiny-en
Cloudflare Workers AI-Workers AI free allocation
Cohere Transcribe Arabic (07-2026)
cohere-transcribe-arabic-07-2026
Cohere25 MBfree evaluation key - 1,000 calls/month shared
Cohere Transcribe (03-2026)
cohere-transcribe-03-2026
Cohere25 MBFree evaluation key — 1,000 calls/month shared; language defaults to English
Voxtral Mini Transcribe 2 (Mistral)
voxtral-mini-2602
Mistral25 MBRenewing Mistral Free-mode monthly usage shared with Studio/API/Vibe; account-specific allowance. Paid overage may apply if enabled.
Whisper Large v3 (OVH)
whisper-large-v3
OVH AI Endpoints25 MBfree · anonymous 2 req/min per IP
Whisper Large v3 Turbo (OVH)
whisper-large-v3-turbo
OVH AI Endpoints25 MBfree · anonymous 2 req/min per IP
Typhoon ASR Realtime
typhoon-asr-realtime
Typhoon-Ongoing free research API (shared rate limits)
Typhoon Isan ASR Realtime
typhoon-isan-asr-realtime
Typhoon-Ongoing free research API (shared rate limits)

Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.

Official limits, checked 20 November 2026
HostWhisper modelsFree limit (official)Best for
GroqWhisper Large v3, Whisper Large v3 Turbo25 MB for each file on the free tier. Audio under 10 seconds counts as 10 secondsSpeed and accuracy
Cloudflare Workers AIWhisper Large v3 Turbo, Whisper, Whisper Tiny English10,000 neurons each day. Large v3 Turbo uses 46.63 neurons for each audio minuteVTT subtitles, about 214 audio minutes each day
OVH AI EndpointsWhisper Large v3, Whisper Large v3 Turbo2 requests per minute for each IP address and model, with no keyFallback with no sign-up

Add Groq and Cloudflare keys together. When Groq sends a 429 error, the router sends the same file to the next host in the chain. OVH is a good last fallback, because its anonymous limit is low. The free limits of each host are separate, so each extra key adds capacity.

Whisper Large v3 or Large v3 Turbo?

Groq publishes a word error rate (WER) for both models. A lower WER means fewer wrong words.

ModelWER on GroqTranslation on GroqChoose it when
whisper-large-v310.3%YesAccuracy matters most, for example names, numbers and accents
whisper-large-v3-turbo12%NoYou need fast results for clear speech

Whisper Tiny English on Cloudflare is small and supports only English. Use it only for short, clear English audio. Both large models are multilingual.

Translation: the FreeLLMAPI docs list only /v1/audio/transcriptions. There is no /v1/audio/translations endpoint. To get English from other languages, transcribe first, then translate the text with a chat model.

Step 1: Install FreeLLMAPI and add keys

FreeLLMAPI is open source. It runs on your computer in Docker. Your provider keys stay on your computer.

Install (macOS or Linux)

curl -fsSL https://freellmapi.co/install.sh | bash

Open the dashboard at http://localhost:3001. Add a Groq key and a Cloudflare key on the Keys page. The Cloudflare key has the form account_id:token. OVH Whisper needs no key. Copy your unified key from the same page.

Step 2: Transcribe with Whisper

Send the file as multipart/form-data. Use model=auto or model=whisper-1 to let the router choose. To pin a model, send its id. The id whisper-large-v3-turbo matches the rows on Groq and OVH, so failover stays on the same model.

curl: pin Whisper Large v3 Turbo

curl http://localhost:3001/v1/audio/transcriptions \
  -H "Authorization: Bearer <unified-key>" \
  -F [email protected] \
  -F model=whisper-large-v3-turbo \
  -F response_format=verbose_json

Python: subtitles from Cloudflare Whisper

from openai import OpenAI

client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")

with open("lecture.mp3", "rb") as audio:
    vtt = client.audio.transcriptions.create(
        model="@cf/openai/whisper-large-v3-turbo",  # Cloudflare makes VTT natively
        file=audio,
        response_format="vtt",
    )
open("lecture.vtt", "w", encoding="utf-8").write(vtt)

Long recordings: compress and split

FreeLLMAPI and the Groq free tier both accept at most 25 MB in one upload. Groq says that Whisper downsamples audio to 16 kHz mono. Do this before upload to make the file much smaller. For long recordings, split the audio into parts.

ffmpeg: 16 kHz mono FLAC parts of 10 minutes

ffmpeg -i podcast.mp3 -ar 16000 -ac 1 -c:a flac -f segment -segment_time 600 part%03d.flac

Python: transcribe the parts in order

import glob
from openai import OpenAI

client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")
texts = []
for path in sorted(glob.glob("part*.flac")):
    with open(path, "rb") as audio:
        extra = {"prompt": texts[-1][-200:]} if texts else {}  # end of the previous part
        r = client.audio.transcriptions.create(model="auto", file=audio, **extra)
    texts.append(r.text)
print("\n".join(texts))

A word can be cut at the border between two parts. Groq suggests short overlaps between parts for exact results. Check each part size before upload, because some audio does not compress well.

Get better Whisper transcripts

  • Set the language: send language with a code such as en, de or ur. Whisper then does not need to detect the language.
  • Use the prompt: send names, product names and special terms in prompt. Whisper uses them as context for spelling.
  • Keep temperature at 0: a higher value gives more varied text, which is rarely useful for transcripts.
  • Clean the audio: remove music and long silences. Record speech close to the microphone.
  • Get timestamps: use response_format=verbose_json. When the provider returns segments, each one has a start and end time. If it has no segments, you get plain JSON with the text only.
  • Check the host: the X-Provider and X-Model response headers show which host and model made the transcript. Log them when you compare quality.

Troubleshooting

ProblemCauseFix
Error 413 file_too_largeThe upload is larger than 25 MBConvert to 16 kHz mono FLAC, or split the file
Words repeat, or text appears in silenceWhisper can write text that was not spoken in long silent partsCut long silences before upload
Names are spelled wrongWhisper does not know the namesSend the names in the prompt field
Error 429 from OVHThe anonymous limit is 2 requests per minuteAdd a Groq or Cloudflare key
Error 400 on srtNo provider makes SRTUse vtt with a Cloudflare Whisper model
Use every one of these through one key. FreeLLMAPI is an open-source, self-hosted router that puts all these free tiers behind a single OpenAI-compatible endpoint and fails over when one is rate-limited. Browse the catalog or go live.

Frequently asked questions

Is there a free Whisper API?

Yes. Groq, Cloudflare Workers AI and OVH AI Endpoints run Whisper free with limits. FreeLLMAPI puts them behind one OpenAI-compatible endpoint with failover.

What is the file size limit of the free Whisper API?

Groq allows 25 MB for each file on its free tier, and FreeLLMAPI accepts uploads up to 25 MB. Convert the audio to 16 kHz mono or split it for longer recordings.

Is Whisper Large v3 Turbo as accurate as Whisper Large v3?

Not quite. Groq lists a word error rate of 12% for Turbo and 10.3% for Large v3. Turbo is faster and costs less on paid plans.

Can the free Whisper API translate audio to English?

FreeLLMAPI has no translation endpoint. Transcribe the audio first, then translate the text with a free chat model.

Does the OpenAI Python SDK work with a free Whisper API?

Yes. Set base_url to http://localhost:3001/v1, then call client.audio.transcriptions.create with model="auto" or a Whisper model id.

Sources

All guides · Browse all free models

Keep reading

All posts →
Provider

Free Gemini API

Google AI Studio's permanent free tier — Gemini and Gemma, no credit card.

Provider

Free DeepSeek API

DeepSeek V3, R1 and V4 served free across multiple providers.

Provider

Free Llama API

Meta's Llama models free across 14 providers, behind one key.