Deepgram Pricing: What Voice Agents and Transcripts Cost

Deepgram pricing starts with $200 in free credit, then usage-based charges: Nova-3 recorded transcription costs $0.0043 per minute and the Standard Voice Agent API costs $0.075 per connected minute on Pay As You Go. That puts 40 one-hour meeting recordings at $10.32 a month, or 2,000 five-minute support calls at $750, before credits, taxes and external services. Start on Pay As You Go; move to Growth when predictable usage or peak concurrency justifies an annual commitment.

Verified against Deepgram's pricing page, October 2026, with the pages checked on October 20. All prices are USD. The workload totals below are calculations from published rates, not measured production results or accuracy tests.

Deepgram Pricing at a Glance

Deepgram, an API provider for turning speech into text, generating speech and running voice agents, lists Pay As You Go, Growth and Enterprise. Growth buys prepaid usage at different rates; it is not an unlimited subscription.

These are the plan names and prices on the live pricing page. The $4,000 Growth floor is about $333.33 per month when annualized, but Deepgram does not advertise that as a monthly payment option. Its advertised savings of up to 20% are not a uniform discount across every model.

Choose on three criteria: your billable units, the amount you can confidently use during the commitment, and simultaneous demand. A lower per-minute rate cannot rescue an annual purchase you barely use.

Deepgram API Pricing: What Gets Metered

Speech-to-text charges for processed audio duration, text-to-speech charges for input characters, and the Voice Agent API charges for WebSocket connection time. A WebSocket is the persistent connection your application keeps open during a conversation. Those units describe different products and cannot be substituted for one another.

For transcription, Deepgram states that it bills by the second. It also charges for each audio channel processed: a stereo recording sent as two separate channels consumes twice the billable duration of a mono recording of the same length. For a managed voice agent, budget the connected session, including quiet time while the connection stays open. Source: billing FAQ and rate tooltips.

The managed agent price includes the listening, language-model processing and speaking stack described by Deepgram. Do not add standalone transcription and speech-generation rates again for components already included in your chosen managed configuration. Bring-your-own configurations need the external provider's bill added separately.

For a phone product, also budget your carrier, phone numbers, application hosting, storage and monitoring. The worked agent total below is the Deepgram subtotal. The voice-agent platform comparison helps place that component inside a broader buying decision.

Deepgram Speech to Text Pricing

Recorded meetings should use the pre-recorded rate card when live transcripts are unnecessary. Streaming means sending audio while the conversation is happening; it has a separate rate card and currently carries several promotional prices.

Deepgram Nova 3 Pricing

Nova-3 Monolingual pre-recorded costs $0.0043/minute on Pay As You Go and $0.0036/minute on Growth. Nova-3 Multilingual pre-recorded costs $0.0052 and $0.0043 respectively. These are base transcription rates; the meeting example below uses the monolingual model. Source: pre-recorded pricing.

For Nova-3 Monolingual streaming, current Pay As You Go/Growth rates are $0.0048/$0.0042 per minute; the page strikes through regular rates of $0.0077/$0.0065. Nova-3 Multilingual streaming is $0.0058/$0.0050, against displayed regular rates of $0.0092/$0.0078. Deepgram labels these limited-time promotions without giving an end date in that section. Use current rates for today's estimate and the displayed regular rates for a budget stress test. Source: streaming pricing.

Deepgram Flux

Flux English costs $0.0065/minute on Pay As You Go and $0.0057/minute on Growth, currently discounted from displayed regular rates of $0.0077 and $0.0065. Flux Multilingual is $0.0078 and $0.0068 per minute respectively. Flux is Deepgram's conversational speech-recognition family, with turn detection for recognizing when a speaker has finished. That is a different purchasing requirement from transcribing a completed recording. Source: Flux rate rows.

Deepgram vs Whisper

Deepgram's hosted Whisper Large is $0.0048/minute on both self-serve plans, so the published Growth rate offers no discount on that row. Custom speech-to-text models require a sales quote. The pricing FAQ also lists the older Enhanced and Base models at $0.99/hour and $0.87/hour on Pay As You Go. Do not assume a legacy model is the budget option. Source: Deepgram pricing.

Price does not establish whether Nova-3, Flux or Whisper will transcribe your callers more accurately. Compare them on representative recordings before switching; this article makes no measured accuracy claim.

Text-to-Speech: Price the Characters

Deepgram text-to-speech is billed per 1,000 input characters, not per minute of output audio. Aura-1 costs $0.0150 on Pay As You Go and $0.0135 on Growth; Aura-2 costs $0.030/$0.027; Flux TTS costs $0.0450/$0.0405. Source: TTS pricing.

At one million input characters, the Pay As You Go subtotal is therefore $15 for Aura-1, $30 for Aura-2 or $45 for Flux TTS. Aura-2's rate is $0.000030 per character. A voice product should count the text it sends for synthesis; converting to a spoken-minute estimate requires an explicit assumption about speaking speed and text length.

The current Flux TTS promotion gives eligible Pay As You Go and Growth projects $1 of bonus credit for each $1 of qualifying paid Flux TTS usage, up to $500 per project, through December 31, 2026. Spending the signup credit does not qualify, and the bonus funds later API usage. It is not an immediate reduction to the listed rate. Source: promotion terms.

Choose a voice for your actual scripts before optimizing character cost. The low-latency text-to-speech API guide covers the wider voice-agent selection problem.

Deepgram Voice Agent API

Standard costs $0.075 per connected minute on Pay As You Go or $0.068 on Growth. The other published configurations change which components Deepgram supplies. BYO means bring your own; LLM means the language model that produces the agent's replies. Source: Voice Agent rate card.

Standard - BYO TTS is $0.065/minute on Pay As You Go and $0.051 on Growth. Custom - BYO LLM + TTS is $0.050 and $0.041. Advanced is $0.163 and $0.146, while Advanced - BYO TTS is $0.122 and $0.110. These all use connection minutes, not just the seconds when the agent speaks.

Custom - BYO LLM needs extra care: the live desktop table displays $0.059/minute in the Pay As You Go column but leaves the Growth value blank. Confirm the applicable rate in the console or with Deepgram before budgeting that configuration. A blank cell is not a free tier or permission to infer another plan's price. Source: current public table.

A BYO configuration only saves money overall if its Deepgram charge plus your external component charges beats the managed alternative at acceptable quality. For another phone-agent budgeting approach, see the GPT-Live-1 phone-agent cost analysis.

Concurrency Limits: The Ceiling Your Monthly Total Hides

Concurrency limits simultaneous work, not how many minutes you can buy. A support line can have modest monthly usage and still exceed its limit during a short burst of calls.

The pricing page lists Pay As You Go at up to 50 speech-to-text REST requests, 150 speech-to-text WSS connections and 5 Deepgram Whisper Cloud requests. Text-to-speech is up to 45 across REST + WSS, Voice Agent up to 45 WSS connections, and Audio Intelligence up to 10 REST requests. REST is the request/response API; WSS is the secure streaming connection.

For Growth, the listed ceilings are 50 speech-to-text REST, 225 speech-to-text WSS, 5 Whisper Cloud, 60 text-to-speech REST + WSS, 60 Voice Agent WSS and 10 Audio Intelligence REST. Enterprise has no numeric ceiling on the pricing page; obtain the agreed concurrency in the quote.

For a meeting recorder, simultaneous file uploads can usually be scheduled. A live support caller needs capacity when the call arrives. That distinction can justify a plan change before a volume discount does.

Worked Monthly Costs: Support Calls and Team Meetings

The same vendor can cost a team about ten dollars for recordings and hundreds for a conversational agent because the products do different work. These examples use the published rates above and explicit workload assumptions.

Support Voice Agent: $750 a Month on Pay As You Go

Assume 2,000 support calls per month, averaging 5 connected minutes, with a peak of 10 simultaneous calls. Use the managed Standard configuration and assume the full connected duration is billable.

2,000 × 5 = 10,000 minutes. At $0.075/minute, that is $750/month, or $0.375 per call. With the entire $200 signup credit still available, the first month's remaining Deepgram charge would be $550. The recurring gross estimate remains $750. Rate source.

On Growth, those minutes consume $680 in credits per month, or $8,160 over a year at $0.068/minute. That is $70/month less at the standard Growth rate, provided the usage is covered by sufficient prepaid credit. Buying only the minimum annual block will not fund the whole year at this volume.

If those minutes instead fall into overage, $0.068 × 1.10 becomes $0.0748/minute. A full 10,000-minute month in overage costs $748 before tax, just $2 below Pay As You Go. Prepaid-credit planning matters more than the headline discount here.

Team Meeting Transcripts: $10.32 a Month

Assume a 20-person team shares 40 one-hour meetings per month. Each meeting produces one mono (single-channel) recording, transcribed once with Nova-3 Monolingual pre-recorded. Attendee count does not multiply the recording duration.

40 × 60 = 2,400 billable minutes. At $0.0043/minute, the monthly subtotal is $10.32, or $0.258 per meeting. Pre-recorded speaker diarization, which labels who spoke when, and Smart Formatting are included on the pricing page. Summaries and other paid analysis are excluded. Rate and feature source.

Process two separate audio channels instead and the subtotal doubles to $20.64. The meaningful multiplier is channels processed, not the number of colleagues in the room.

Growth would consume $8.64/month for the mono workload, a $20.16 annual rate saving against Pay As You Go. That does not justify a $4,000 prepayment. Stay on Pay As You Go unless other workloads or capacity requirements change the decision.

When Growth Pays for Itself

Compare the cash commitment with the usage you can defend, then check how much of the prepaid block you will consume. Do not treat $4,000 as an extra platform fee on top of every minute: it buys credits.

For a single-product workload, ignoring signup credits, taxes, overages and rate changes, a $4,000 minimum Growth purchase first matches Pay As You Go spending at $4,000 divided by the Pay As You Go unit rate. This is a cash crossover, not the point at which every prepaid credit has been used.

For Standard agents, that is about 53,334 minutes a year, or 4,445 a month. Fully consuming $4,000 at the Growth rate takes about 58,824 minutes a year. For Nova-3 Monolingual pre-recorded, the cash crossover is about 930,233 minutes a year, or 77,520 a month; full consumption at the Growth rate takes about 1,111,112 minutes a year. These are computed thresholds from the listed rates, with annual volumes rounded up.

A mixed workload needs a weighted forecast across its actual models and add-ons. Choose Growth when the prepay is justified by that forecast or required capacity. Choose Enterprise when the public limits or deployment options do not fit, and require a concrete quote.

Extra Charges, Credit Expiry and Renewal Terms

The main sources of a larger bill are paid features, more processed channels, open agent connections and credit-management settings. Keep those in the estimate you hand to a founder.

Redaction adds $0.0020/minute on Pay As You Go or $0.0017 on Growth; Keyterm Prompting adds $0.0013/$0.0012; Entity Detection adds $0.0017 on either plan, for both recorded and streaming transcription. Streaming speaker diarization displays $0.0020/minute in the Pay As You Go column, with no separate Growth value shown. Pre-recorded diarization and Smart Formatting are included. Source: add-on rows.

Summarization has its own token meter: $0.0003 per 1,000 input tokens and $0.0006 per 1,000 output tokens on Pay As You Go; Growth lists $0.00024/$0.00048. Tokens are the units the text model processes. A transcription-minute budget does not automatically include that charge. Source: Audio Intelligence pricing.

Growth overages carry the 10% premium plus tax, billed weekly on Monday for the previous week's usage. A valid payment card is required for overages; exhausted credits without a card can produce a 402 response. Deepgram says early renewal avoids the premium. Pay As You Go auto-load defaults to charging $100 when the balance reaches $10 and can be disabled in settings. Source: billing policies.

The pricing page says the signup credit remains available until used. That does not establish unconditional lifetime credit retention for every plan. Deepgram's terms make Growth an automatically renewing annual subscription, allow credit expiry on account or subscription closure, and provide no general obligation to roll over unused credits. They also distinguish the subscription payment from the rates at which credits are consumed. Confirm the expiry and rollover treatment of your purchased Growth block before relying on unused balance for another year. Source: terms, sections 5.3 and 5.4.

Deepgram Alternatives: Two Direct Pricing Comparisons

Compare the same recorded minutes and required features before declaring a winner. These are base transcription prices, without an accuracy ranking or an assumption that the feature bundles are identical.

OpenAI's GPT-Transcribe, a speech-to-text model, lists $0.0045/minute, making the same 2,400 recorded minutes $10.80, versus Deepgram Nova-3 Monolingual's $10.32. Source: OpenAI pricing.

AssemblyAI's Universal-3.5 Pro, a recorded-audio transcription model, lists $0.21/hour, equivalent to $0.0035/minute and $8.40 for 2,400 minutes, versus Deepgram's $10.32 before any paid feature differences. Source: AssemblyAI pricing.

For this meeting workload, the difference is small enough that transcript usability and required features should decide the shortlist. For a large production system, repeat the calculation with your own volume and configuration.

Frequently Asked Questions

Is Deepgram free to use?

New accounts receive $200 in signup credit, with no card required. It is a finite balance, not a monthly free allowance. At the selected $0.0043 recorded rate it buys about 46,511 minutes before add-ons; at Standard agent pricing it buys about 2,666 minutes. A one-off project that stays within the credit need never pay. An ongoing application eventually does. Source.

How much does Deepgram cost?

The price depends on the API and model. Pay As You Go starts with the signup credit and then charges usage; Growth starts at $4,000 per year in prepaid credits; Enterprise requires a quote. The examples above cost $750/month for the support agent and $10.32/month for mono meeting transcription before credits and excluded services. Source.

How does Deepgram work?

Your application sends recorded or live audio for transcription, sends text for speech generation, or maintains a connection to the Voice Agent API for conversation. These are developer APIs; the quoted rates pay for processing rather than a per-seat meeting-app subscription. Source.

What is Deepgram's latest model?

There is no single model covering every product. The current rate card names Nova-3 and Flux for speech recognition, plus Flux TTS, Aura-2 and Aura-1 for speech generation. Choose by endpoint and task rather than treating one family as a universal replacement. Source.

Who does Deepgram compete with?

OpenAI and AssemblyAI are two speech-to-text alternatives with direct published prices, compared above. For a complete phone agent, compare the full voice stack and carrier costs rather than only transcription.

Is Deepgram better than Whisper?

This pricing analysis does not establish an accuracy winner. Deepgram itself sells hosted Whisper Large alongside its own speech-recognition models. Evaluate representative audio and the features you need; a lower rate alone does not establish better output. Source.

Is Deepgram a big company?

Deepgram announced a $130 million Series C at a $1.3 billion valuation in January 2026. That establishes funding scale, not the capacity guaranteed to your account. Your relevant limit is the contracted API concurrency. Source: Deepgram announcement.

Is Deepgram profitable?

The official pricing, company and funding pages checked for this article do not establish profitability. A funding round or valuation does not answer that question.

Who owns Deepgram AI?

The service is provided by Deepgram, Inc.; its company page names Scott Stephenson as CEO and co-founder. Its funding announcement lists investors, but does not disclose a complete ownership breakdown. Company page, funding announcement.

Does Deepgram offer a student discount?

No student-specific discount is published on the pricing page checked for this article. The publicly listed starting offer is the $200 signup credit. Source.

What is Deepgram's refund policy?

The pricing page's Pay As You Go policy allows refund requests for unused purchased credits bought within the previous 30 days; older purchases are non-refundable. Do not assume this is a general refund promise for an annual Growth commitment or promotional credits. Source: refund policy.

Has Deepgram pricing changed?

The current streaming table explicitly distinguishes promotional prices from struck-through regular prices. Those are the verified comparison points here; the table does not establish when every historical rate changed or when the streaming promotion ends. Source.

Before choosing a plan, total next month's connected calls, recorded channel-minutes and synthesized characters, then check the busiest expected period. Use the AI Business Workflow Audit Checklist to turn that forecast into a build decision.