Deepgram STT tiers per hour vs Sume: Nova-3, Whisper, Flux

Deepgram's cheapest speech-to-text rate, Nova-3 pre-recorded mono at $0.0043 per minute, is $0.258 per hour, against $0.60 per hour for Sume STT 1.0. Its most expensive listed rate here, Flux streaming for English at $0.0065 per minute, is $0.39 per hour. Every Deepgram rate on the page read on 2026-10-05 is therefore lower than Sume's $0.01 per minute, by between 35 and 57 percent.

Five Deepgram rates and Sume

Deepgram's pricing page lists streaming and pre-recorded rates by model. The table picks five of them: three pre-recorded (Nova-3 mono, Whisper Large and Nova-3 multilingual) and two streaming (Nova-3 mono and Flux for English). It multiplies each per-minute rate by 60 for an hourly figure and by 6,000 for 100 hours. The last column is how much more Sume STT 1.0 would cost for the same 100 hours.

Where the gap comes from

Sume STT 1.0 is the provider list of $0.008 per minute times 1.25, so $0.01 per minute, shown on the API pricing page. Deepgram's cheapest pre-recorded rate is below the provider list Sume starts from, so even before the margin the two are not the same model at the same price. The Sume figure is for a managed request type with one key and one balance, which is worth something if transcription is a step in a longer pipeline and worth nothing if it is the only thing you buy. Be clear with yourself about which of those you are.

What the table leaves out

Deepgram's page lists add-ons and different rates for other products, and the table uses mono audio. Streaming and pre-recorded work differently, and the mono assumption matters if your files are multichannel, because the page may price channels separately: the streaming rates suit live calls and the pre-recorded ones suit files. Sume STT 1.0 is a request-based, file-style service with a 10-minute limit per request, so the closest comparison is Deepgram's pre-recorded rows, which cost $0.258 to $0.312 an hour.

• Pre-recorded Nova-3 mono: 57 percent below Sume.
• Pre-recorded multilingual: 48 percent below Sume.
• Flux streaming: 35 percent below Sume.

Monthly volumes

Here are three monthly volumes for the nearest like-for-like pair: Deepgram Nova-3 pre-recorded mono and Sume STT 1.0. Each cell is hours times the hourly rate, with no free credits or plan fees.

The spread grows linearly, so the question is where it becomes larger than the cost of running a second vendor. For most teams that is somewhere in the hundreds of hours a month.

When Sume is still the sensible choice

At 10 hours of audio a month the difference is $2.10 against the Flux streaming rate and $3.42 against Nova-3 pre-recorded, which no one picks a vendor for. At 1,000 hours it is $342 against the cheapest Deepgram rate. If your pipeline already generates video, speech and captions on Sume, the same balance covers the transcript. If transcription is the whole product, price it on Deepgram's page and run a 10-hour sample through each service first, using audio like your real traffic: the same languages, the same noise level and the same number of speakers. An accuracy gap of a few percentage points costs more in correction time than the per-hour gap saves.

Rates change, so read Deepgram's page the day you decide.