Cloudflare Workers AI Whisper vs Sume STT: cost of 1,000 hours

One thousand hours of audio costs $30 on Cloudflare Workers AI Whisper and $600 on Sume STT 1.0, which is a 20 times difference. Cloudflare's pricing page, read on 2026-10-05, lists Whisper and whisper-large-v3-turbo at $0.0005 per audio minute. Sume's rate is $0.01 per audio minute. This post does the arithmetic and says what the gap does and does not tell you.

The numbers

A thousand hours is 60,000 minutes. At $0.0005 per minute that is $30. At Sume's $0.01 per minute it is $600. The Sume rate is the provider list of $0.008 per minute multiplied by 1.25, in line with how every Sume model is billed.

What a 20 times gap is made of

Two things are in it. First, Cloudflare publishes its own per-minute rate for Whisper on its platform. Sume bills a hosted STT model at the provider list rate times 1.25. Second, the products differ: Cloudflare Workers AI is a compute platform where you write the code that calls the model, while Sume STT is one request type among video, audio and image, with one balance and one usage log.

This post does not compare accuracy. Whisper and the model behind Sume STT are different models, and you should run a sample of your own audio through both before you let price decide.

Where Cloudflare's price needs a second look

The pricing page prices the model, not the whole job. Anything you run around it, such as splitting files, queueing, retries and storing transcripts, is yours to build and pay for on Cloudflare's side. On Sume the request limits are explicit: STT 1.0 takes up to 10 minutes of audio per request, so a long file has to be cut before submitting, and each request is billed for its own minutes.

• Cloudflare: $0.0005 per audio minute for Whisper and whisper-large-v3-turbo.
• Sume: $0.01 per audio minute, maximum 10 minutes per request.
• Both: bill per minute of audio, not per minute of processing time.

A worked pipeline

Say a podcast network transcribes 40 hours of new audio a month, then trims highlights and burns captions. Transcription is 2,400 minutes, which is $24 on Sume and $1.20 on Cloudflare. Trimming is $0.02 per call and captions are $0.20 per job on Sume, so for 40 episodes with five highlight clips each, that is 200 trims at $4 and 200 caption jobs at $40, for $44 on top of the transcription.

In that pipeline the STT line item is real money relative to Cloudflare, but it is small compared with the video work around it. That is the usual shape: the cheapest STT rate matters most when STT is nearly the whole bill, and matters least when it is one line among several, each of which you would otherwise buy from a different vendor with its own key and invoice.

When the gap matters

At 10 hours of audio a month, the difference is $5.70. Nobody picks a stack over that. At 1,000 hours it is $570, which is worth an afternoon of engineering. At 100,000 hours it is $57,000 and it is the whole decision.

If transcription is a small step before something Sume does anyway, such as captioning or cutting a video, the integration cost on the Sume side is zero because the same key and balance cover it. If transcription is the product, a bare-compute price is hard to beat.

Check before you commit

Rates move. Read Cloudflare's current pricing page and the Sume API pricing page on the day you decide, and use the usage dashboard to compare against your real minutes after a trial week.