Deepgram Pricing Explained (2026): STT, Aura & Voice Agent Costs

Deepgram charges by usage, not by seat. Nova-3 speech-to-text costs $0.0043/min pre-recorded and $0.0077/min streaming, Aura-2 text-to-speech is $0.030 per 1,000 characters, and new accounts start with $200 in free credit (~45,000 minutes). There's no monthly minimum on pay-as-you-go — you pay for what you process.

Quick answer

Deepgram has two paid tiers plus a free credit: Pay As You Go (no commitment, published per-unit rates) and Growth (a committed-spend plan, ~$4,000/year, that discounts those rates). Enterprise contracts exist above that with custom terms, self-hosting, and compliance. For nearly everyone starting out, pay-as-you-go is the right choice.

Speech-to-text rates

Speech-to-text is metered per minute of audio, and the streaming vs pre-recorded split is the main variable.

Model	Pay As You Go	Growth
Nova-3 (pre-recorded / batch)	$0.0043/min	discounted with commit
Nova-3 (streaming / real-time)	$0.0077/min	$0.0065/min
Nova-3 multilingual (streaming)	slightly higher	discounted

To put that in human terms: streaming Nova-3 is about $0.46 per hour of audio, and batch is closer to $0.26 per hour. That's cheap enough that transcription is rarely the biggest line item in a voice product.

Aura text-to-speech rates

Text-to-speech is billed per 1,000 characters, not per minute — an important difference if you're used to STT math.

Model	Pay As You Go	Growth
Aura-2 (enterprise voices)	$0.030 / 1,000 chars	$0.027 / 1,000 chars
Aura-1	$0.015 / 1,000 chars	$0.0135 / 1,000 chars

As a rule of thumb, 1,000 characters is roughly 150 spoken words, or about a minute of speech. So Aura-2 costs around $0.03 per minute of generated audio — cheaper per minute than most premium TTS, though the voices are less expressive than ElevenLabs.

Growth plan and free credit

Every new account gets $200 in free credit — up to ~45,000 minutes of Nova transcription — with no card required to start. That's enough to prototype an entire voice agent before spending anything.

The Growth plan is a committed-spend tier that starts around $4,000/year and cuts per-unit rates by roughly 15-20%. The math only works once your usage is high enough that the discount clears the commitment — if you're spending less than a few hundred dollars a month, stay on pay-as-you-go.

What a real bill looks like

Say you run a voice agent handling 1,000 five-minute calls a month — 5,000 minutes of two-way audio. Streaming STT at $0.0077/min is about $38.50. If the agent speaks roughly half the time, TTS at ~$0.03/min for the agent's ~2,500 minutes is about $75. Add your LLM cost separately, and Deepgram's share of a 5,000-call month lands near $115 — before the free credit and before any Growth discount.

The takeaway: at small-to-mid scale, Deepgram is a minor cost. It becomes material only at very high volume, which is exactly when the Growth plan starts to pay off.

Is it worth it?

For real-time voice, yes. You're getting sub-300ms streaming latency, class-leading accuracy, and infrastructure that scales to thousands of concurrent streams, at prices that stay small until you're genuinely big. The only reason to look elsewhere on cost is if you already run your own GPUs and can self-host open-source Whisper — and even then you give up latency, support, and managed scaling.

For the full capability rundown, read our Deepgram review, and if you're building an agent, see Deepgram vs ElevenLabs and the best AI voice tools roundup.