Deepgram vs AssemblyAI
Both sell transcription by usage, but they count differently and charge for extras differently. Deepgram bills per minute, with a $200 starting credit and prepaid Growth credits from $4,000 a year for lower rates. AssemblyAI bills per hour, lists a lower base rate for its older Universal-2 model, and offers a free tier measured in hours of audio. Speaker labelling shows the gap most clearly, at about $0.12 an hour on Deepgram's list and $0.02 an hour on AssemblyAI's. Country of establishment is absent from both vendors' pages as read, so it plays no part here.
Vendors can pay for more visibility on this website, for example a higher place in a ranking or more room on a page. Commercial relationships can also affect which products we feature and how prominently. Every vendor link is marked the same, whatever the arrangement behind it. How that works.
you want one per-minute rate with no minimum, a starting credit that does not expire, and a prepaid route to cheaper rates once volume is predictable. Extras like speaker labelling arrive as separate line items, and you can live with that.
you want a free tier sized in hours, a cheaper base rate on the older model, and low-cost speaker labelling. You are also content to read the per-hour list and to pay about 10 percent more for in-region US or EU processing.
Deepgram and AssemblyAI side by side
| Deepgram | AssemblyAI | |
|---|---|---|
| Our speech-to-text apis rank | #4 of 12 | #2 of 12 |
| Established in | Not stated by the vendor | Not stated by the vendor |
| List price, 12 November 2026 | Billed per second at per-minute rates; Pay as you go starts with a $200 credit, Growth prepaid from $4,000 a year, Enterprise by contract | Per audio hour by model: Universal-2 $0.15, Universal-3.5 Pro $0.21; free tier up to 185 recorded hours; volume pricing through sales |
| Free tier | — | — |
| Open source | — | — |
| Self-hostable | Yes | — |
| Built for | Engineering teams processing large volumes of audio, or needing on-premise | Developers who want to start on credits and pick analysis features one by one |
| Not for | Office users who want to upload a recording and read it | Teams that need one flat all-in hourly price |
Both are reviewed in full in our speech-to-text apis ranking: Deepgram and AssemblyAI.
Where they differ
Deepgram 3 · AssemblyAI 3 · open 0- 01
Minutes against hours on the invoice
Deepgram's table quotes Nova-3 monolingual at $0.0077 per minute on Pay as you go and $0.0065 on Growth, billed per second without block rounding. That is about $0.46 and $0.39 an hour. AssemblyAI quotes Universal-2 at $0.15 per hour and Universal-3.5 Pro at $0.21 per hour. On base rates alone, AssemblyAI's list is lower, though the two models differ in languages and accuracy claims, which neither list lets us compare. Check the model you would actually run before reading either column as the real bill.
Edge: AssemblyAI
- 02
What the first free audio buys
Deepgram's Pay as you go plan carries a $200 credit, needs no card, and the page says it has no expiry. At the Nova-3 monolingual rate that is roughly 430 hours before add-ons. AssemblyAI's FAQ states a free tier of up to 185 hours of recorded transcription and up to 333 hours of streaming, also without a card. Deepgram's credit stretches further for recorded audio, while AssemblyAI separates the two modes in advance, which makes live-audio testing easier to plan.
Edge: Deepgram
- 03
Speaker labels and other extras
On Deepgram, speaker diarization is $0.0020 per minute, redaction $0.0020, keyterm prompting $0.0013 and entity detection $0.0017, while smart formatting is included. AssemblyAI lists speaker diarization at $0.02 per hour, keyterms prompting at $0.05 per hour on Universal-3.5 Pro and included on Universal-2, and medical mode at $0.15 per hour. Diarization is the clearest gap, about six times cheaper per hour at AssemblyAI. Both stack extras on top of the base rate, so the add-on list belongs in any cost model.
Edge: AssemblyAI
- 04
Committing money for a better rate
Deepgram offers Growth: prepaid credits for the year, redeemed against actual usage, from $4,000 per year and with savings of up to 20 percent according to the page. That helps a buyer who can forecast volume and ties up cash for a year. AssemblyAI's page promises pay-as-you-go with no commitments and routes volume discounts to a conversation with the sales team, with no published prepaid tier. A team that cannot forecast its volume may prefer that flexibility to a year of prepaid credit.
Edge: Deepgram
- 05
Live audio and voice agents
AssemblyAI lists Universal-Streaming at $0.15 per hour for English, plus a realtime Universal-3.6 Pro model at $0.45 per hour for voice agents, and marks two newer realtime models as coming soon. Deepgram's table covers Flux STT at $0.0077 per minute, and a separate Voice Agent API from $0.050 per minute that bundles recognition, a language model and speech output. Deepgram's bundled agent pricing is the more fully published of the two, and AssemblyAI's streaming list is the cheaper for plain English transcription.
Edge: Deepgram
- 06
Where audio is processed
AssemblyAI states that its listed prices are for global routing and that in-region US or EU processing costs 10 percent more, so a buyer with a residency requirement should model that uplift. The Deepgram page we read mentions concurrency that differs by region on Growth, with limits for North America, Europe, Australia and India, but it gives no residency price. Neither page states a country of establishment for the company. Residency is therefore a question to put to each sales team in writing.
Edge: AssemblyAI
When neither fits
If the audio must never leave your own hardware, or you want a model under an open licence, neither suits. Vosk is open source and runs offline, and Soniox bundles speaker detection into one hourly rate.
Deepgram or AssemblyAI: questions
3 answersWhich of the two charges less for speaker labels on recorded files?
AssemblyAI lists speaker diarization at $0.02 per hour. Deepgram lists it at $0.0020 per minute, which works out to about $0.12 per hour. Both are charged on top of the base transcription rate.
Does either one expire a starting balance?
Deepgram's page says its Pay as you go plan has no expiration and no minimums, and the $200 credit needs no card.
AssemblyAI's FAQ describes a free tier counted in hours rather than money. No expiry date for the AssemblyAI tier appears in the text we saw.
Can a small team prepay for cheaper audio at either vendor?
Deepgram sells Growth, prepaid credits from $4,000 per year, with savings of up to 20 percent on the page. AssemblyAI describes pay-as-you-go with no commitments and handles volume pricing through sales. So prepaying is a published option only at Deepgram.
Sources
Checked 12 November 2026- Deepgram pricing page: Nova-3 per-minute rates, $200 credit, Growth from $4,000 a year, add-on rates, Voice Agent rates.
- AssemblyAI pricing page: hourly model rates, add-on features, free tier hours, 10 percent in-region uplift, realtime models.