Deepgram Nova-3 $0.0048 and Flux $0.0065 a minute vs Sume STT $0.01

Deepgram lists streaming Nova-3 monolingual at $0.0048 a minute and Flux English at $0.0065, both marked promotional on its pricing page, which is $0.288 and $0.39 an hour. Sume's speech-to-text is $0.01 a minute, or $0.60 an hour. The gap is 2.1 and 1.5 times, and it buys a different shape: Sume takes a public audio URL as a job and returns word timings, it does not accept a live stream.

Deepgram figures are from its pricing page, read 2026-10-03; the hourly figures are my arithmetic.

Prices per minute and per hour

Deepgram's page calls both rates promotional, so check whether they have changed before budgeting beyond this week.

What differs besides price

Sume STT takes a public HTTPS audio URL and an optional language code; omit it to auto-detect. Duration is capped at 600 seconds per job, and word timings are always returned. Add the segmentation option and you also get sentence segments. See the API reference.

A one-hour recording therefore becomes six jobs of 10 minutes, each costing $0.10, and your code joins the word timings by adding each part's start offset. The walk-through is in splitting long recordings.

Monthly scenarios

Multiply the hourly rates by the audio you actually process. The gap grows with volume, and so does the number of Sume jobs: 40 hours is 240 jobs of 10 minutes.

Which to use

For a live caption or a spoken assistant, a streaming API is the only option; Sume does not offer one. For recorded files, the extra $0.21 to $0.31 an hour at Sume is the price of one job-based API shared with the video, image and audio steps. Eight hours a month is $4.80 at Sume against $2.30 on promotional Nova-3, a $2.50 difference.

• Live captions and agents: Deepgram or another streaming vendor.
• Recorded webinars, calls and shows: either, depending on whether you need the rest of your pipeline in one API.
• Check the promotional flag on Deepgram's rates before locking a budget.