The free tier, in full
- 15 transcripts per IP address per 24 hours
- Up to 75 minutes of audio per file
- Up to 180 minutes of audio a day — resets at midnight UTC; the fast lane has no daily cap
- OpenAI Whisper large-v3-turbo — the fast top-tier checkpoint
- Timestamped paragraphs on every result
- Download as .txt, .srt, or .vtt — no account required
- 100 languages with browser-locale + metadata pre-detection
- YouTube, podcasts, audio URLs — any public source our extraction pipeline can reach
- Direct file upload — drag-and-drop or browse, up to 100 MB per file
- 3-link batch, speaker diarisation, translate-to-English — all included
- No signup, no email, no Chrome extension
- Audio deleted within 24 hours, transcripts not retained
The fast lane
- Roughly 69× real time — a 60-minute episode comes back in under a minute
- No queue, ever — paid jobs run off-box and never wait behind anyone
- Files up to 3 hours, against 75 minutes on the free lane
- 600 minutes of audio, charged by real length and rounded up to the minute
- Minutes never expire — no monthly reset, no clock running on the balance you bought
- Refunded in full if a transcript fails — you are never billed for nothing
- No account — an access link, saved in your browser and emailed to you
- One-time payment. There is also $5/month for unlimited minutes if you transcribe constantly
What it does not change: the transcript. Both lanes run Whisper large-v3-turbo, so paying buys the same text sooner — not better text. Speaker identification, and word-level timestamps need our own pipeline, so those jobs stay on the free lane at free-lane speed even once you have paid.
Why speed is the only thing we sell
We looked at a week of real usage before deciding what to charge for. Of 146 completed transcriptions between 19 and 25 August, 69% took more than five minutes and 21% took more than half an hour; the longest single wait was 74 minutes. Waiting, not the daily allowance, is what people actually run out of — so waiting is what the paid lane removes, and it is the only thing we ask money for.
This is also why we are not selling a "pro" transcript. Charging more for better accuracy would mean deliberately running a weaker model on the free tier, and we are not doing that. The free lane gets the same Whisper checkpoint the paid lane does. The difference is the machine it runs on and whether you wait in line for it.
The free tier is not a trial with a countdown. In August we raised its per-file ceiling from 30 to 75 minutes because the failure logs showed people bouncing off it — and we did that before there was any paid tier to protect.
Frequently asked
Is shrybe really free?
The free tier really is free. No credit card, no email, no signup. The limits are 15 transcripts per IP address per 24 hours, 75 minutes of audio per file, and 180 minutes of audio a day. Those caps exist because compute costs are real, and we would rather cap the free tier than wrap it in ads.
What am I actually paying for?
Time, and nothing else. The free lane runs on our own machines and moves at roughly real time — a 60-minute episode takes about 65 minutes, longer if someone is ahead of you in the queue. The fast lane runs the same audio off-box at roughly 69 times real time, so that same episode comes back in under a minute, and it never queues. The transcript is identical: same Whisper large-v3-turbo model on both lanes.
Is the transcript better if I pay?
No, and we will not pretend otherwise. Both lanes run Whisper large-v3-turbo, so a paid transcript is the same text you would have got for free after waiting. If accuracy is your problem rather than speed, paying will not fix it.
What does the fast lane not do?
Speaker identification and word-level timestamps need our own pipeline, so jobs using them stay on the free lane at free-lane speed even if you have paid. Translate-to-English is not one of them any more — it runs on the finished text, so it works on either lane. Batch is still three links. The 15-transcripts-a-day cap applies on both lanes: the monthly plan is unlimited in minutes of audio, not in requests per day. The free lane also has a 180-minute-a-day audio budget; the fast lane does not.
How are minutes counted?
By the actual length of the audio, rounded up to the next minute. A 19-second clip costs one minute; a 62-minute podcast costs 62. If a transcript fails, the minutes go straight back — you are never charged for something we did not deliver. Minutes bought once do not expire: there is no monthly reset and no deadline to use them by, so a balance you top up in March is still there in December.
Do I need an account?
No. Paying creates an access token, not a login. It is saved in the browser you paid from, and emailed to you as a link so you can use it on another device. There is no password to forget and nothing to cancel except a subscription, if you chose one.
Will the free tier get worse now that there is a paid one?
No. That is the bet most freemium products lose, and we are not making it. The free tier as it stands — 15 per day, 75 minutes per file, 180 minutes of audio a day, Whisper large-v3-turbo, timestamps, diarisation, SRT/VTT export — is the floor, not a trial. We raised the free cap from 30 to 75 minutes in August because the data said people needed it, and that was before there was anything to sell.
Try the free tier.
YouTube · Podcasts · Audio files · Captions · Summaries. Same form, same 15-per-day cap.