Deepgram Pricing Explained: Speech, Voice and Voice Agent Rates

Deepgram bills speech-to-text per second: Nova-3 streaming is $0.0077 a minute on pay as you go for one language, with add-ons extra. Aura-2 voices cost $0.030 per 1,000 characters, and the Voice Agent API is $0.075 a minute on Standard. Dograh, which you can self-host or run on Dograh Cloud, runs Deepgram's models with your own key.

Key Takeaways

• Speech-to-text, voices and the agent each use a different meter and unit.
• Streaming add-ons such as redaction and diarization can add about 70% per minute.
• The $200 credit does not expire, and Growth starts at $4,000 a year prepaid.

This post is part of our guide to Deepgram alternatives, where each Deepgram product is priced on its own meter. Here we go through each rate and add-on in turn, then the plans and limits, and end with what each number buys next to Dograh.

How Deepgram bills

Deepgram uses a different meter for each product, so read each rate in its own unit. On Deepgram's pricing page, speech-to-text is priced per minute of audio and billed per second, with no rounding up to blocks. Multichannel audio is billed on the total duration across channels.

Text-to-speech is priced per 1,000 characters of input text. The Voice Agent API is priced per minute and billed on WebSocket connection time, so a call that sits open costs the same as one where people talk.

Each product has two self-serve rates. Pay as you go has no minimum. Growth is a prepaid plan with lower rates, covered below. Enterprise prices are agreed with sales.

Speech-to-text and add-on prices

Streaming speech-to-text, the mode a live call uses, costs $0.0077 a minute for one language on pay as you go. The full card for streaming and recorded audio:

Per hour, one-language streaming comes to $0.46 on pay as you go and $0.39 on Growth. Custom-trained models are priced by sales.

Add-ons are charged per minute on top of the base rate. On pay as you go, redaction is $0.0020, keyterm prompting $0.0013, entity detection $0.0017 and speaker diarization $0.0020 on streaming audio. Smart formatting is included, and diarization is included on recorded audio.

They add up. By our arithmetic at those rates, 10,000 minutes of Nova-3 streaming on pay as you go is $77, and $130 with every streaming add-on except entity detection switched on.

Keyterm prompting tells the model which words to listen for, such as product names. Deepgram lists it as a paid add-on, and Dograh's Dictionary uses it, so check your Deepgram usage report. Our post on using a custom dictionary for industry jargon and names explains when the Dictionary is worth it.

Open Source Alternative to Vapi / Retell

Self-hosted voice agent platform — no per-minute fees

dograh-hq/dograh

Star on GitHub

Text-to-speech prices

Voices are priced per 1,000 characters of text. Flux TTS is $0.0450 per 1,000 characters on both plans. Aura-2 is $0.030 on pay as you go and $0.027 on Growth, and the older Aura-1 is $0.0150 and $0.0135.

The bill follows what the agent says. A short, direct script costs less to voice than a chatty one, and the caller's side of the call costs nothing on this meter.

Voice Agent API prices

The Voice Agent API is "one rate covering speech-to-text, a managed language model, and text-to-speech", and the tier depends on which language model you pick.

Per hour, Standard is $4.50 on pay as you go. Deepgram's model list puts smaller models such as gpt-5-mini, gpt-4o-mini, claude-haiku-4-5 and gemini-2.5-flash on Standard. Larger ones, such as gpt-5, gpt-4.1, claude-sonnet-4-5 and gemini-3-pro-preview, are Advanced.

Bringing your own language model or voice lowers Deepgram's rate, but you then pay that provider directly. The phone line is not in any tier. You pay your phone provider for it, and you run the bridge between the call and the WebSocket.

Plans, free credit and limits

A new account gets a $200 credit with no card needed, and it does not expire. Deepgram says it covers roughly 43,000 minutes of Nova-3.

Growth is prepaid, at "$4,000+ per year in prepaid credits", and Deepgram says it saves up to 20%. Usage past your prepaid credit is charged at the Growth rate plus 10%, weekly. Without a card on file, requests are refused once credits run out.

Concurrency is capped by plan. Streaming speech-to-text allows 150 connections at once on pay as you go and 225 on Growth. The Voice Agent API on Growth allows up to 60 at once in North America and 45 in Europe, Australia and India. Enterprise limits are set to your peak traffic, and requests over a limit are queued or rejected.

Self-hosting is an Enterprise feature. Deepgram's self-hosting docs say the containers need an Enterprise plan and NVIDIA GPUs, and report usage to Deepgram's license server without sending audio. Enterprise rates go down with committed volume.

One more setting touches your data. Deepgram's Model Improvement Program is voluntary, and each request can opt out with mip_opt_out=true.

What each number buys, with Dograh

The rates above buy different things, so line them up by what they include before comparing.

Dograh replaces the Voice Agent API. Add your Deepgram key in Model Configurations, and Dograh can keep Deepgram as its speech layer: Nova-3 or Flux for speech-to-text and Aura-2 voices, billed by Deepgram. Self-hosted Dograh has no per-minute platform fee (servers, model keys and phone minutes still cost). Dograh Cloud has a 1¢-a-minute platform fee, and Dograh's managed models are about $0.07 a minute for most models, billed separately from it.

So for a team on the Voice Agent API, the fair comparison is one bundle against a platform fee plus the models it chooses. Our post on BYOK vs managed models explains how to keep that model spend in check, and Dograh vs Deepgram compares the two agents feature by feature.