Deepgram pricing 2026: STT, TTS, and agent costs

Deepgram prices almost everything by the minute of audio, which makes it one of the easier speech bills to predict. The catch is in the details: which model, whether the rate on the page is a promotion, how many channels your audio has, and which add-ons you switch on.

Here is the short version, read from Deepgram’s pricing page in October 2026. Streaming speech to text costs $0.0048 to $0.0078 per minute on Pay As You Go, with several of those rates marked as limited-time promotions; the regular rate for Nova-3 and Flux English is $0.0077. Text to speech costs $0.015 to $0.045 per 1,000 characters. The Voice Agent API costs $0.075 per minute on its Standard tier. New accounts start with a $200 credit that doesn’t expire.

We build Cartesia, which competes with Deepgram on speech to text, text to speech and voice agents, so we have an interest here. Every Deepgram number below comes from Deepgram’s own pages, and our prices are at the end so you can check both.

Deepgram plans

Pay As You Go needs no credit card to start and no minimum. Credits auto-reload by default ($100 when the balance hits $10), which you can turn off in the console. Purchased credits are refundable for 30 days if unused.

Concurrency is where many live applications hit a wall before cost does. Pay As You Go allows up to 150 streaming speech-to-text connections and 45 text-to-speech or voice agent connections at once; Growth raises those to 225 and 60. Past the limit, requests may queue or be rejected.

Speech-to-text pricing

These are per-minute rates in October 2026. “Promo” is the current price Deepgram shows next to a crossed-out regular price.

Flux is Deepgram’s model for voice agents, with turn detection built in. Nova-3 is its general transcription model. Older models are still available at their old rates: Nova-2 streaming at $0.35 per hour, Enhanced at $0.99 and Base at $0.87.

The promotional rates have no end date on the page. If you are budgeting a year, use the regular column; it is what you pay when the promotion ends.

Three billing rules change the total more than the model choice does:

• Billing is per second. Deepgram doesn’t round up to the minute.
• Multichannel audio bills per channel. A 10-minute stereo call is 20 billed minutes. Contact centers that record agent and caller on separate channels pay double.
• Add-ons bill per minute on top. Redaction is $0.0020, entity detection $0.0017, keyterm prompting $0.0013 and streaming diarization $0.0020 on Pay As You Go. Smart formatting is included, and so is diarization on pre-recorded audio.

Text-to-speech pricing

At roughly 1,000 characters per minute of speech, Flux TTS works out to about 4.5 cents per minute of audio. Deepgram is running a credit match on Flux TTS until December 31, 2026: every dollar spent adds a dollar of credit, up to $500.

Voice Agent API pricing

The Voice Agent API bundles listening, an LLM and speech into one WebSocket and bills by connection time. Which tier you pay depends on the LLM. Small models such as GPT-4o mini, Claude Haiku 4.5 and Gemini Flash models are Standard; larger ones such as GPT-5 and Claude Sonnet are Advanced, per Deepgram’s LLM list.

Two things sit outside these rates. Phone calls need a carrier such as Twilio, which bills you separately. And the bring-your-own tiers are cheaper only on Deepgram’s line of the invoice: you still pay the LLM or TTS provider you bring.

Because billing runs on connection time, an agent that holds the socket open through hold music or a long silence pays for those minutes too.

What a real workload costs

Two examples at Pay As You Go list prices.

Transcribing 100,000 minutes of two-channel call recordings a month. That is 200,000 billed minutes. On Nova-3 pre-recorded at $0.0043, the bill is $860. Adding redaction ($0.0020 per minute) brings it to $1,260. On Growth, both rates drop by roughly 15 to 20%.

A voice agent handling 10,000 call minutes a month. On the Standard tier, that is $750 to Deepgram, plus your carrier. If the agent needs a larger model, the Advanced tier makes it $1,630.

Where Cartesia fits

If you are pricing Deepgram for a voice agent, three numbers matter more than the per-minute rate: how many words the agent hears correctly, how natural it sounds, and what a whole call costs.

• Speech to text: on the Artificial Analysis streaming benchmark in October 2026, Ink-2 has a 3.4% word error rate, against 6.6% for Nova-3 Realtime and 7.4% for Flux. Every misheard account number or name is a turn the agent has to repair.
• Text to speech: in Artificial Analysis blind listening tests, Sonic 3.6 scores 1284 against 1054 for Flux TTS (October 2026).
• Voice agents: calls on Cartesia Managed Agents cost $0.06 per minute, against $0.075 for Deepgram’s Standard tier. For a limited time, LLM usage is included for agents built in the Cartesia playground. The 10,000-minute agent above costs $600 on Cartesia; a Cartesia phone number adds $0.014 per minute.

Our Cartesia vs Deepgram comparison has the latency measurements, and the speech-to-text model guide covers the other streaming models worth testing.

If you are transcribing recorded audio in bulk and don’t need a live agent, Deepgram’s pre-recorded rates are built for that job, and per-second billing keeps short files cheap.

Cartesia pricing, for comparison

Cartesia’s plans bill speech to text and text to speech from one pool of credits: 1 credit per character of text to speech, and 3 credits per second of Ink-2 audio.

Voice agents cost $0.06 per minute, drawn from the prepaid agent balance first. Unused credits roll over up to twice the monthly amount. Cartesia also deploys in your VPC, on-prem, or air-gapped.

Try Ink-2 and Sonic in the playground, or run the same audio through both APIs before you commit to either bill.

Related guides

• ElevenLabs pricing 2026
• Best speech-to-text models
• What is a voice agent?
• AI voice agent platforms