Deepgram Review 2026: Speech-to-Text at $0.0043 a Minute — Plus the Add-On Math

What does Deepgram do?

Deepgram sells voice infrastructure: speech-to-text, text-to-speech, and a combined Voice Agent API, all billed by usage. Its flagship transcription model is Nova-3, offered in Monolingual and Multilingual versions for both streaming (live audio over a websocket) and pre-recorded files. A newer Flux line is tuned for conversational agents, with built-in turn detection and interruption handling so a voice bot can tell when a caller has finished speaking.

Transcripts come with the extras production apps actually need: speaker diarization (who spoke when), smart formatting for punctuation, casing, dates and currency, keyterm prompting to boost product names and jargon, and PII redaction. An Audio Intelligence layer adds summarization, sentiment analysis, intent recognition and topic detection on top of the transcript. Nova models support 45+ languages, and Deepgram also hosts Whisper Large ($0.0048/min) if you prefer that model behind the same API.

For voice agents, the Voice Agent API bundles speech-to-text, a language model connection and text-to-speech (its Aura-2 voices cost $0.030 per 1,000 characters standalone) into one websocket, priced per minute of connection time.

How much does Deepgram cost?

There is no subscription in the usual sense. Pay As You Go has no minimums and starts with a $200 free credit — no credit card required. Growth prepays $4,000 or more per year in credits and cuts per-minute rates by up to about 20%. Rates below are from Deepgram’s own pricing page, checked October 2026.

Do the add-on math before you budget. Speaker diarization is included on pre-recorded transcription but costs $0.0020/min on streaming; redaction adds $0.0020/min and keyterm prompting $0.0013/min. A fully loaded streaming minute can cost close to double the headline rate — still cheap at roughly 29 cents per hour of audio for the base Nova-3 streaming model, but not the number on the banner.

What are Deepgram’s strengths?

• Genuinely cheap at scale: pre-recorded Nova-3 works out to about $0.26 per hour of audio before add-ons.
• One vendor for the whole voice stack: transcription, speech synthesis and the agent layer share one account, one bill and one set of credentials.
• Built for live audio: Flux models handle turn detection and interruptions, which is where home-grown voice agents usually fall apart.
• Compliance boxes ticked: SOC 2 Type II, HIPAA (with a BAA for Enterprise customers), GDPR with an EU endpoint, and PCI compliance.
• A real free start: $200 of credit with no card is enough to transcribe thousands of minutes and find out whether accuracy holds up on your audio.

What are Deepgram’s weaknesses?

• API only: there is no meeting bot, no transcript editor, no friendly dashboard workflow for non-developers.
• Add-on pricing: diarization, redaction and keyterm prompting are separate line items on streaming, so the real per-minute cost needs a spreadsheet.
• Growth is a commitment: the better rates require $4,000+ prepaid per year — money wasted if your volume never shows up.
• Per-minute metering cuts both ways: the Voice Agent API bills by websocket connection time, so idle connections cost money.

Who is Deepgram for?

Developers and product teams building voice agents, call analytics, captioning or media transcription into their own software — especially anyone whose volume makes per-seat transcription apps expensive. It is a poor fit for individuals who just want meeting notes: tools like Otter.ai or Sonix give you an app today, while Deepgram gives you the engine to build one.

Frequently asked questions

Is Deepgram free?

Not permanently. New accounts get a $200 credit with no credit card required; after that, everything is pay-as-you-go per minute or per character.

How much does Deepgram cost per hour?

Nova-3 pre-recorded transcription at $0.0043/min works out to about $0.26 per hour of audio, before add-ons like streaming diarization or redaction.

Does Deepgram do text-to-speech as well?

Yes. Aura-2 costs $0.030 per 1,000 characters on Pay As You Go, and the newer Flux TTS costs $0.045 per 1,000 characters.

Deepgram vs Whisper — which should I use?

You may not have to choose: Deepgram hosts Whisper Large at $0.0048/min alongside its own Nova-3 models, so you can test both behind one API.

Verdict

Deepgram is easy to recommend to the audience it is actually built for. The pricing is published, the free credit is generous, the model range covers batch transcription through to real-time agents, and the compliance story holds up for enterprise use. Just go in with your eyes open about add-ons and the Growth prepay — and if you need an app rather than an API, look elsewhere first.

Prices checked October 2026. This is a research-based review; we do not claim hands-on testing.