Why does the Whisper API reject files over 25 MB? and what to do with a long file instead.
The OpenAI Whisper API refuses any file above 25 MB per request, which at 128 kbps is only about 26 minutes of MP3, so a long recording has to be compressed or cut into pieces before it is accepted. WhipScribe takes one file up to 10 hours or 5 GB (2 GB before the first pack) at about 2 minutes per audio hour, the first 20 minutes are free, and a 2-hour file uses 120 of the 500 minutes in the $6 pack.
25 MB, translated
What a 25 MB request cap means for an hour-long file.
The line in minutes
Size is bitrate times length. At 128 kbps, 25 MB holds roughly 26 minutes; at 64 kbps about 52 minutes; at 32 kbps about an hour and three quarters. Video is worse, because the picture takes most of the bytes. The cap is a rule of that API on request size, not a judgement on your audio.
What chunking costs you
Getting around it means writing code that cuts the audio into pieces, ideally at silences, sends each one, then adds the right offset to every timestamp and joins the text. Words spoken across a cut get broken, and every piece has to be retried on its own when one fails.
One file instead of two dozen
Ten hours of 128 kbps audio is about 576 MB: twenty-four pieces under a 25 MB cap, or one upload here. Up to 5 GB goes through the upload page, files over 2 GB start once a pack is on the account, and the transcript keeps one clock from the first second to the last.
More than a transcript
Each file comes back with a brief, chapters with timestamps, the names, numbers and dates said, the questions the recording answers and a chat over the text. Exports are TXT, DOCX, SRT, VTT and JSON, so the JSON slots into whatever you were going to build anyway.
Pieces or one file
Working around a request cap is a project in itself.
✗ Chunking for a 25 MB cap
The work a long file needs before that API accepts it.
- Compress to a low bitrate first
- Cut the audio at silences in code
- Send and retry every piece
- Offset each set of timestamps
- Repair sentences split at the cuts
- Merge the pieces into one document
✓ One upload here
The same long file, sent once.
- Original file up to 5 GB
- Up to 10 hours in one piece
- Timestamps on a single clock
- No sentences broken at cuts
- A brief and chapters for the whole
- JSON and SRT ready to use
Sample output
Minute 70, no seam in sight.
Illustrative lines invented for this page, not from a real recording, from a long technical talk that would have needed several pieces under a 25 MB cap.
Export
One transcript. Five clean formats.
Every pack exports all five — no format is held back for a higher tier.
Plain text
De-ummed paragraphs. Ready to paste.
SRT captions
Timed to the speech. Every video editor reads this.
WebVTT
HTML5 player + YouTube uploads.
Show notes
Formatted with chapters and pull-quotes.
Machine-readable
Timestamped text for your own tools.
Pricing
Honest pricing, no surprises.
A pack from $6 (500 minutes) opens the full transcript. After that, no subscription and no auto-renew — every tier is a one-time minute pack, and credits never expire. Buy once, use the minutes whenever you need them.
Casual pack
$6/500 min
8 hours of audio — $0.72 an hour. One-time — credits never expire.
- 500 minutes of audio
- All export formats
- AI chat over the whole transcript
- Executive summary & key quote
Best value
$12/2,000 min
33 hours of audio — $0.36 an hour. Credits never expire.
- 2,000 minutes of audio
- Summary, chapters and key quotes
- All export formats
- AI chat over the whole transcript
Bulk pack
$24/5,000 min
83 hours of audio — $0.29 an hour. Credits never expire.
- 5,000 minutes of audio
- Summary, chapters and key quotes
- API access for batch pipelines
- All export formats
FAQ
Questions about the 25 MB limit.
Why does the Whisper API reject files over 25 MB?
Its file upload is capped at 25 MB per request, so anything larger is turned away before transcription begins. For speech that works out to about 26 minutes at 128 kbps, which is why long meetings, lectures and interviews hit it first.
Can I compress my file to fit under 25 MB?
Sometimes. Mono speech at 32 to 64 kbps stays understandable and fits roughly 52 to 105 minutes. Pushing lower starts to blur consonants, and recordings over two hours still have to be cut into pieces.
Can I send a 3-hour file to WhipScribe in one go?
Yes. One file can run up to 10 hours or 5 GB, so three hours, even as video, goes up whole. It takes about 6 minutes to process and uses 180 of the 500 minutes in the $6 pack.
Is there an API for long files here?
API access comes with the $24 pack of 5,000 minutes, as listed on /pricing. If you only need transcripts of the files you have, the website does it without writing any code.
What does a month of long recordings cost?
Add up the running time. Thirty 1-hour calls are 1,800 minutes on the $12 pack of 2,000; forty 2-hour lectures, 4,800 minutes, fit the $24 pack of 5,000. Credits never expire, so a quiet month costs nothing.
Where is my audio processed?
Recordings are transcribed on our own servers and are never used for training; the audio is not passed to a third-party AI service, and you can delete a recording at any time.
Related
Related pages on file limits.
Refused at 25 MB? Send the whole file here instead.
Upload my fileOperated by Neugence Technology Pvt. Ltd. · contact@whipscribe.com · Security · Privacy · Terms