Deepgram
Fast, accurate speech-to-text to power high-volume voice apps, for teams happy to build on a developer API.
Paid link, we may earn a commission. How this works.
Scored on the same voice-agent rubric as the full platforms, so a building block like this scores low on the axes it does not address. Read its value score against its job.
See how it stacks up · Full rankings →The cheap all-in-one engine. Deepgram folds speech-to-text, the language model and its Aura-2 voice into one API at about $0.075 a minute, among the lowest going. No built-in phone line though, so you add Twilio, and only seven languages.
About $0.08 to 0.18 for a minute of conversation, once the phone line and the AI are added in.
That's roughly $4.80–10.80 an hour. Plans: $0/mo (Pay as you go).
Pricing
Show the cost breakdown
| What the platform charges to run the agent, before the phone line and the AI usage are added on. | $0.08 /min |
|---|---|
| The step that turns what the caller says out loud into text the AI can read. | — |
| The AI 'brain' that reads what the caller said and works out what to say back. | — |
| The step that turns the AI's written reply back into a spoken voice. | — |
| The phone line itself: the service that connects the call to a real phone number. Usually billed on top of the platform. | $0.01 /min |
| The total you actually pay for one minute of conversation once every piece is added up: the platform, the AI, the voice and the phone line. | $0.08–0.18 /min |
The Standard Voice Agent rate is $0.075/min and bundles speech-to-text, the language model and Aura-2 text-to-speech, billed on websocket connection time. Telephony is not included: you bring Twilio (about $0.014/min, a third-party cost), which is why the all-in here adds a telephony line. Cheaper tiers let you bring your own LLM or TTS, and the page now publishes the full BYO grid: Standard BYO TTS $0.065/min PAYG ($0.051 Growth), Custom BYO LLM $0.059, Custom BYO LLM+TTS $0.050 ($0.041 Growth), Advanced BYO TTS $0.122 ($0.110); you then pay the external model or voice on top. An Advanced Voice Agent tier (premium models/voices) sits above Standard at about $0.163/min. Aura-2 narration is $0.030 per 1,000 characters (Aura-1 is half that). Streaming speech-to-text LIST rates: Nova-3 mono $0.0077/min ($0.0092 multilingual) and Flux English $0.0077/min; at the 2026-07-23 capture the page struck those through under a 'limited-time promotional rates on streaming' banner, charging $0.0048 (Nova-3 mono), $0.0058 (multilingual) and $0.0065 (Flux English) instead, with Flux Multilingual at $0.0078 carrying no strikethrough. A footnote adds that listed rates opt in to the Model Improvement Program (data-sharing), the same pattern as Speechmatics' discount toggle, so treat the promo figures as conditional twice over; we store list per house policy. The voice library has no single canonical total: about 90 Aura-2 voices across 7 languages (the stored 91 was not reproducible). SOC 2 Type 1 and Type 2, a HIPAA Business Associate Agreement on request, GDPR and PCI are stated on Deepgram's trust page, which now also names CCPA and an EU endpoint (api.eu.deepgram.com).
Every plan in one place: the monthly fee, what each one includes, and the features it unlocks. Anything beyond a plan's allowance, or on a pay-as-you-go tier, is billed at the per-minute rate above. A blank in the features means the vendor's plan page does not state it for that plan, not that it is unavailable.
| Pay as you go | Growth | Enterprise | |
|---|---|---|---|
| Price | Free | — | Custom |
| Included | Pay per use | Pay per use | — |
| Plan notes | $200 free credit, no minimums or expiry | Discounted per-minute rates; pricing page now shows a $4K+/year floor with up to 20% pre-paid savings (2026-07-11) | Custom volume pricing |
| What each plan unlocks | |||
| API access | Yes | Yes | — |
| Concurrent calls | Up to 45 (Voice Agent) | Up to 60 (Voice Agent) | — |
| Priority support | Community + Discord | Community + Discord | Custom volume / deployment |
- Pay as you go FreePay per use
$200 free credit, no minimums or expiry
- API access
- Yes
- Concurrent calls
- Up to 45 (Voice Agent)
- Priority support
- Community + Discord
- Growth —Pay per use
Discounted per-minute rates; pricing page now shows a $4K+/year floor with up to 20% pre-paid savings (2026-07-11)
- API access
- Yes
- Concurrent calls
- Up to 60 (Voice Agent)
- Priority support
- Community + Discord
- Enterprise Custom—
Custom volume pricing
- API access
- —
- Concurrent calls
- —
- Priority support
- Custom volume / deployment
Prices in USD as set by the vendor · last checked 2026-07-30 · vendor pricing →
Slide your expected monthly volume to see roughly what Deepgram would cost.
A rough estimate from Deepgram's sourced rates, not a quote. Always confirm on the vendor's own pricing page before you commit.
At a glance
- Speech-to-text
- Deepgram Nova-3, Deepgram Flux
- Text-to-speech
- Deepgram Aura-2
- Languages
- en, es, de, fr, nl, it, ja
- Integrations
- Twilio, OpenAI, Anthropic, ElevenLabs (BYO TTS), Cartesia (BYO TTS), AWS Polly, Docker / Kubernetes self-host, Native SDKs (JS/Python/Go/.NET)
Compliance
Our full take
Deepgram started as a speech-to-text company, and it shows in the pricing. The Voice Agent API takes the three pieces a phone agent needs, the speech-to-text, the language model and the voice, and bills them as one number instead of three separate meters. On the Standard tier that number is about $0.075 a minute, which is among the lowest all-in rates of any serious platform.
The trade you make for that price is the phone line. Deepgram does not bring its own telephony, so you wire up Twilio yourself and pay Twilio separately, usually around $0.014 a minute on top. For a developer that is a small job. For a non-technical buyer who wants to plug in and go, it is real setup work, and worth knowing before you compare the headline rate to an all-in platform like Vapi or Retell.
If you want to spend even less, the cheaper tiers let you bring your own language model or your own voice engine (ElevenLabs, Cartesia and AWS Polly all plug straight in), which drops Deepgram’s cut to about $0.041 a minute on the Growth plan. You then pay the outside model or voice on top, so the saving is smaller than it first looks, but the flexibility is genuine.
The voices come from Deepgram’s own Aura-2 engine, with the older Aura-1 still available at half the per-character price for anything where top quality is not the point. Narration runs about $0.030 per 1,000 characters on Aura-2, competitive but not the cheapest. The real limit is reach: seven languages and no documented voice cloning, so if you need to sound local in twenty markets, or clone a specific brand voice, this is not the tool.
One pricing habit to watch before you budget from the rate card. As of our 23 July 2026 capture, the streaming transcription rates are labelled limited-time promotional, with the list price struck through beside them: Nova-3 runs $0.0048 a minute on promo against a list of $0.0077, and a footnote says the listed rates also assume you opt in to Deepgram’s Model Improvement Program, which shares your audio to train its models. We store the list rates on this page. The promo is real money while it lasts, but a budget built on it inherits both an expiry date and a data-sharing condition.
Where Deepgram is unusually strong is the compliance paperwork. Its own trust page states SOC 2 Type 1 and Type 2, a HIPAA Business Associate Agreement on request, plus GDPR and PCI, and now names CCPA and an EU endpoint too. That is more than several flashier competitors can show in writing, and it matters if you are in healthcare or finance. The company underneath it is well capitalised: a $130 million Series C at a $1.3 billion valuation announced 13 January 2026 (which also folded in OfOne, a drive-through voice AI), and a July 2026 deal putting Nova-3 on-device on Qualcomm’s Snapdragon X chips. Neither changes what you pay today, but both matter if you are betting a production stack on the supplier still being here in five years.
The 1 to 10 scores on this page are an editorial preview, our provisional read to get the framework in place, not a measured result. We have not run Deepgram through our own call tests yet, so there is no Voxrater latency figure here. The pricing, voice and compliance detail is sourced from Deepgram’s pricing, product and trust pages, first captured 2026-05-31 and most recently re-captured 2026-07-23.
Deepgram compared
Our in-depth pieces that put Deepgram side by side with the field, with the sourced numbers and a clear pick.
Alternatives to Deepgram
Other platforms that overlap with Deepgram on the same kind of work, ranked by how many capabilities they share, then by cheaper all-in cost per minute. Compare any of them side by side on the compare page.
Tracking Deepgram? Get the next test result
We re-test and re-price the platforms we cover. Join the list and the next dated update lands in your inbox.
Newsletter launching soon.
Sources
- Re-verified 2026-07-30 (screenshot in evidence/): pricing page re-captured and diffed against the 2026-07-23 capture; no change to any rate, tier or plan figure. · captured 2026-07-30
- Deepgram pricing re-captured 2026-07-23 (screenshot in evidence/): Voice Agent $0.075/min Standard and $0.163/min Advanced unchanged; streaming STT table shows a 'limited-time promotional rates on streaming' banner with strikethrough list prices (Nova-3 mono $0.0077/min list vs $0.0048 promo, $0.0065/$0.0042 on Growth; Nova-3 multilingual $0.0092 vs $0.0058; Flux English $0.0077 vs $0.0065; Flux Multilingual $0.0078 with no strikethrough); a footnote states listed rates opt in to the Model Improvement Program (data-sharing); full BYO grid published (Standard BYO TTS $0.065/$0.051, Custom BYO LLM $0.059, Custom BYO LLM+TTS $0.050/$0.041, Advanced BYO TTS $0.122/$0.110); Aura-2 $0.030/1k and Aura-1 $0.015/1k unchanged; streaming add-ons published (redaction $0.0020/min, keyterm prompting $0.0013/min, diarization $0.0020/min). · captured 2026-07-23
- Deepgram press release (2026-01-13): $130M Series C at a $1.3B valuation led by AVP, with Twilio, ServiceNow Ventures, SAP and Citi Ventures among new backers; acquisition of OfOne (voice AI for quick-service restaurants) announced in the same release. · captured 2026-07-23
- Deepgram announcement (dateline San Francisco, 2026-07-21): Nova-3 optimised to run on-device on the Qualcomm Hexagon NPU in the Snapdragon X Series platform, real-time multilingual transcription without a cloud round-trip; developer details via deepgram.com/partners/qualcomm. · captured 2026-07-23
- Re-verified 2026-07-11: Voice Agent $0.075/$0.163/min and BYO $0.041/min unchanged; Growth tier now shows $4K+/year with up to 20% pre-paid savings; Flux English $0.0065/min and new Flux Multilingual $0.0078/min listed; $200 free credit unchanged. · captured 2026-07-11
- Deepgram pricing re-captured 2026-06-15: Standard Voice Agent $0.075/min, Advanced ~$0.163/min; Nova-3 + Flux STT. Aura-2 voice library counts ~90 across 7 languages (no single canonical total; the stored 91 was not reproducible). · captured 2026-06-15
- Deepgram pricing page re-captured 2026-06-02 for the quarterly re-verification; pricing reviewed against the live page (screenshot in evidence/). · captured 2026-06-02
- Deepgram plan page: per-plan features (API, Voice Agent concurrency, support tiers) · captured 2026-05-31
- Deepgram pricing page: Voice Agent Standard $0.075/min, Aura-2 $0.030/1k chars, Nova-3 $0.0048/min · captured 2026-05-31
- Voice Agent API product page: unified STT+LLM+TTS, function calling, bring-your-own LLM/TTS · captured 2026-05-31
- Deepgram trust page: SOC 2 Type 1 and Type 2, HIPAA BAA on request, GDPR, PCI · captured 2026-05-31
- GA announcement: Nova-3 STT, Aura-2 TTS, sub-200ms TTFB claim, barge-in, external TTS providers · captured 2026-05-31
- TTS models docs: 91 voices, 7 languages, Aura-2 and Aura-1 naming · captured 2026-05-31
- Twilio integration guide: telephony is brought in via Twilio Media Streams · captured 2026-05-31