Speech-to-text price per hour, ranked: nine rates and Sume STT
Nine STT rates in dollars per audio hour, from Cloudflare's $0.03 to Sume STT 1.0 at $0.60, each read from the vendor's own page with the conversion shown.
Ranked by dollars per audio hour, Sume STT 1.0 at $0.60 is the highest of the nine rates below, and Cloudflare's Whisper at $0.03 is the lowest. Every rate comes from the vendor's own page read on 2026-10-05, converted to a per-hour figure so they can be compared. This is a price list, not a quality ranking, and it mixes batch and streaming products.
The ranking
Where a vendor lists per-minute prices the table multiplies by 60. Deepgram's $0.0043 per minute becomes $0.258 per hour, Amazon's $0.006 becomes $0.36, Cloudflare's $0.0005 becomes $0.03, and Sume's $0.01 becomes $0.60. The last column shows what 100 hours would cost at that rate.
| Rank | Service | Listed unit | Per hour | 100 hours |
|---|---|---|---|---|
| 1 | Cloudflare Whisper | $0.0005/min | $0.030 | $3.00 |
| 2 | Soniox async | $0.10/hr | $0.100 | $10.00 |
| 3 | Speechmatics Melia 1 | $0.12/hr | $0.120 | $12.00 |
| 4 | Speechmatics Standard | $0.23/hr | $0.230 | $23.00 |
| 5 | Deepgram Nova-3 pre-recorded, mono | $0.0043/min | $0.258 | $25.80 |
| 6 | Amazon Transcribe standard batch | $0.006/min | $0.360 | $36.00 |
| 7 | Speechmatics Enhanced | $0.38/hr | $0.380 | $38.00 |
| 8 | Microsoft MAI-Transcribe-2-Streaming (intro price) | $0.54/hr | $0.540 | $54.00 |
| 9 | Sume STT 1.0 | $0.01/min | $0.600 | $60.00 |
What is not like for like
The Microsoft rate is for a streaming model and the page calls it an introductory price through year-end, so it may not hold. Deepgram lists several products and the table uses its pre-recorded mono Nova-3 rate; its streaming and multilingual rates differ. Soniox lists a separate real-time rate. Amazon's rate is the standard batch rate in US East.
Features such as diarization, language count and add-ons are priced separately at several vendors. Speechmatics, for example, lists translation as a paid add-on. None of that is in the per-hour column.
Where Sume's $0.60 comes from
Sume STT 1.0 bills the provider list of $0.008 per audio minute times 1.25, giving $0.01 per minute, and caps one request at 10 minutes of audio. See the Sume API pricing page. The gap to the next highest rate, Microsoft's $0.54, is six cents per hour; the gap to Deepgram is about $0.34 per hour.
- Versus Cloudflare: Sume is 20 times the price.
- Versus Speechmatics Melia 1: Sume is 5 times the price.
- Versus Microsoft's intro rate: Sume is about 11 percent higher.
Budgets at three volumes
The spread looks small per hour and large per year. Here are four of the rates at 100, 1,000 and 10,000 hours of audio. Each cell is hours times the per-hour rate from the ranking above, with no discounts, free tiers or taxes applied.
| Audio hours | Cloudflare Whisper | Speechmatics Melia 1 | Deepgram Nova-3 | Sume STT 1.0 |
|---|---|---|---|---|
| 100 | $3.00 | $12.00 | $25.80 | $60.00 |
| 1,000 | $30.00 | $120.00 | $258.00 | $600.00 |
| 10,000 | $300.00 | $1,200.00 | $2,580.00 | $6,000.00 |
When the cheapest rate is the right one
If transcription is the product and you process thousands of hours, take the lowest rate that passes your accuracy test. At 10,000 hours a month the spread between $0.03 and $0.60 is $5,700.
If transcription is one step in a pipeline you already run on Sume, using the same key and the same balance avoids a second integration, and the extra cost is $0.60 per hour you actually transcribe. That is $6 for 10 hours. Weigh that against the engineering time of a second vendor: a separate account, key, invoice and failure mode to monitor. Under a few hundred hours a month, the single integration usually costs less than the rate gap saves.
How to keep this list honest
Rates change, and several of these pages say so. Re-read the source before you quote a number: Cloudflare, Soniox, Speechmatics, Deepgram, Amazon and Microsoft AI. For your own Sume spend, the usage dashboard shows what each request billed.
Sources
Related posts
More in Comparisons
- Strands Decider 2B vs Cloudflare Clef for gating agent runs
Both return typed answers (yes/no, choice, score) with confidence. Compare what each vendor page says, and what stays your job when a Sume run costs money.
- 3 days of Stripe retries vs 10 Sume attempts 30 seconds apart
Stripe retries live webhooks for up to 3 days; Sume tries 10 times, 30 seconds apart, about 4.5 minutes. After that, poll the job or call redeliver.
- Suno free plan: 7 trial downloads, no commercial rights for ads
Suno's free plan gives 7 lifetime trial downloads for personal use and no commercial rights. What that means for an ad, and the $0.125 per track route on Sume.
- Tavus CVI has six layers; a Sume avatar video is one job
Tavus CVI chains perception, turn-taking, speech, an LLM, TTS and rendering. A Sume talking-video is one request with a script. Which parts you own in each.
Written by Sume