Deepgram Pricing 2026: Speech-to-Text Cost Per Minute & Hour
Deepgram speech-to-text pricing starts at $0.0043 per minute for pre-recorded Nova-3 Monolingual, or about $0.26 per hour of audio. For live streaming, Nova-3 Monolingual is currently $0.0048 per minute, or about $0.29 per hour, under Deepgram’s limited-time promotional pricing.
The useful number is not just the advertised price per minute. Streaming versus pre-recorded processing, multilingual models, speaker diarisation, redaction, keyterm prompting, multiple audio channels, and live-agent connection time can all change the production bill. This page converts the current rates into hourly and monthly costs so you can budget an actual workload rather than compare isolated API prices.
Rates below were checked on 24 August 2026 against Deepgram’s live pricing page. Several streaming prices are explicitly promotional, so we show both the current rate and Deepgram’s displayed regular rate when available.
This page covers cost only. For accuracy, latency, diarisation quality, model selection and product evaluation, read our Deepgram review.
Deepgram pricing: the numbers you probably need first
| Workload | Current Pay-As-You-Go rate | Cost per hour | Best budgeting assumption |
|---|---|---|---|
| Nova-3 Monolingual pre-recorded | $0.0043/min | $0.258/hr | About $0.26 per audio hour |
| Nova-3 Multilingual pre-recorded | $0.0052/min | $0.312/hr | About $0.31 per audio hour |
| Nova-3 Monolingual streaming | $0.0048/min promotional | $0.288/hr | About $0.29 per live audio hour at the current rate |
| Nova-3 Multilingual streaming | $0.0058/min promotional | $0.348/hr | About $0.35 per live audio hour at the current rate |
| Flux English streaming | $0.0065/min promotional | $0.390/hr | About $0.39 per live audio hour |
| Flux Multilingual streaming | $0.0078/min | $0.468/hr | About $0.47 per live audio hour |
If you are transcribing existing files, start your budget with the pre-recorded column rather than the streaming price. The difference looks small per minute, but pre-recorded processing also includes speaker diarisation, whereas streaming diarisation is a paid add-on.
New accounts currently receive $200 of free credit with no credit card required. You can start with Deepgram’s free credit before committing to production volume.
Current Deepgram speech-to-text pricing matrix
Deepgram separates live streaming from pre-recorded transcription. That split should be the first decision in any cost model. Flux is a streaming model for conversational voice applications, while Nova-3 supports both general streaming and pre-recorded transcription.
Streaming speech-to-text pricing
| Model | Pay As You Go now | Displayed regular PAYG rate | Growth now | Current PAYG cost/hour |
|---|---|---|---|---|
| Flux English | $0.0065/min | $0.0077/min | $0.0057/min | $0.390 |
| Flux Multilingual | $0.0078/min | No separate promotional rate shown | $0.0068/min | $0.468 |
| Nova-3 Monolingual | $0.0048/min | $0.0077/min | $0.0042/min | $0.288 |
| Nova-3 Multilingual | $0.0058/min | $0.0092/min | $0.0050/min | $0.348 |
| Custom | Contact sales | Custom | Contact sales | Custom |
The current Nova-3 streaming discounts are substantial. Nova-3 Monolingual is about 38% below the regular rate Deepgram displays alongside it, while Nova-3 Multilingual is about 37% below its displayed regular rate. Deepgram labels these as limited-time promotional rates but does not currently state an end date for the STT promotion.
That makes the regular-rate column important for production forecasting. A service with a narrow margin should not assume $0.0048 per minute will remain available indefinitely just because that is today’s checkout price.
Pre-recorded speech-to-text pricing
| Model | Pay As You Go | PAYG cost/hour | Growth | Growth cost/hour |
|---|---|---|---|---|
| Nova-3 Monolingual | $0.0043/min | $0.258 | $0.0036/min | $0.216 |
| Nova-3 Multilingual | $0.0052/min | $0.312 | $0.0043/min | $0.258 |
| Whisper Large | $0.0048/min | $0.288 | $0.0048/min | $0.288 |
| Custom | Contact sales | Custom | Contact sales | Custom |
For a straightforward uploaded-file workflow, Nova-3 Monolingual is therefore cheaper than its current streaming equivalent before any extras are considered. It also has another cost advantage: speaker diarisation is included for pre-recorded transcription.
What Deepgram actually costs at realistic monthly volumes
Per-minute pricing hides how low the basic transcription bill can be. The more useful calculation is to multiply the rate by the number of audio hours your application processes each month.
| Monthly audio | Nova-3 Mono pre-recorded | Nova-3 Mono streaming now | Nova-3 Multi pre-recorded | Nova-3 Multi streaming now |
|---|---|---|---|---|
| 1 hour | $0.26 | $0.29 | $0.31 | $0.35 |
| 10 hours | $2.58 | $2.88 | $3.12 | $3.48 |
| 100 hours | $25.80 | $28.80 | $31.20 | $34.80 |
| 1,000 hours | $258 | $288 | $312 | $348 |
| 10,000 hours | $2,580 | $2,880 | $3,120 | $3,480 |
Those are base transcription costs only. A recurring mistake in speech API comparisons is treating “$0.0048 per minute” as the cost of the finished workflow. For batch transcription, cost per processed audio hour is a sensible starting unit. For a live product, cost per successfully completed call or session is usually more informative because add-ons, concurrency, retries and the surrounding voice stack start to matter.
Streaming versus pre-recorded pricing changes more than the base rate
The strongest reason to separate these two modes is speaker diarisation. Deepgram includes diarisation in its current pre-recorded pricing. In streaming, it costs an additional $0.0020 per minute on Pay As You Go.
On the current $0.0048 Nova-3 Monolingual streaming rate, diarisation alone increases the speech-to-text line item by roughly 42%. Add redaction as well, and the rate becomes $0.0088 per minute, or about $0.53 per hour.
| Example workflow | Effective PAYG rate | Cost/hour | Cost for 1,000 hours |
|---|---|---|---|
| Pre-recorded Nova-3 Mono | $0.0043/min | $0.258 | $258 |
| Pre-recorded Nova-3 Mono + diarisation | $0.0043/min | $0.258 | $258 |
| Pre-recorded Nova-3 Mono + redaction + keyterm prompting | $0.0076/min | $0.456 | $456 |
| Streaming Nova-3 Mono + diarisation | $0.0068/min | $0.408 | $408 |
| Streaming Nova-3 Mono + diarisation + redaction | $0.0088/min | $0.528 | $528 |
| Streaming Nova-3 Mono + diarisation + redaction + keyterm prompting | $0.0101/min | $0.606 | $606 |
This is why a recorded call archive and a live call assistant should not share the same cost assumption even if both use Nova-3. The underlying transcript requirement may look similar, but the billable feature stack is different.
Deepgram add-on pricing can become a large part of the STT bill
| Feature | Streaming PAYG | Pre-recorded PAYG | Cost effect |
|---|---|---|---|
| Redaction | $0.0020/min | $0.0020/min | Adds $0.12 per processed hour |
| Keyterm prompting | $0.0013/min | $0.0013/min | Adds $0.078 per processed hour |
| Smart formatting | Included | Included | No separate charge |
| Entity detection | $0.0017/min | $0.0017/min | Adds $0.102 per processed hour |
| Speaker diarisation | $0.0020/min | Included | Meaningful difference between live and batch economics |
The practical lesson is to estimate the output you need before choosing the cheapest base model. If a support transcript must contain speaker labels, remove sensitive data and recognise a product-specific vocabulary, all three requirements belong in the original cost model.
Example: Deepgram bills for real workloads
These examples make the gap between headline pricing and workflow pricing clearer. They use current Pay-As-You-Go rates and assume the stated audio duration is actually processed.
| Workload | Assumption | Monthly calculation | Estimated cost |
|---|---|---|---|
| Podcast transcription | 50 hours, Nova-3 Mono pre-recorded, diarisation included | 3,000 min x $0.0043 | $12.90 |
| Meeting archive | 100 hours, Nova-3 Mono pre-recorded with diarisation | 6,000 min x $0.0043 | $25.80 |
| Recorded support calls | 300 hours, Nova-3 Mono + redaction | 18,000 min x $0.0063 | $113.40 |
| Large call archive | 1,000 hours, Nova-3 Mono + redaction + keyterm prompting | 60,000 min x $0.0076 | $456 |
| Live support transcription | 1,000 hours, Nova-3 Mono + diarisation + redaction + keyterm prompting | 60,000 min x $0.0101 | $606 |
| Large batch transcription | 10,000 hours, Nova-3 Mono pre-recorded | 600,000 min x $0.0043 | $2,580 |
The absolute difference between the models is tiny, at 10 or 50 hours per month. At that scale, choosing a model because it saves a few dollars is usually poor optimisation if it creates extra correction work. At thousands of hours, routing, add-ons and contracted pricing become much more relevant.
The $200 Deepgram free credit is large enough for a real benchmark
Deepgram currently gives new Pay-As-You-Go accounts $200 in credit. At the base rates above, that is considerably more than a token demo allowance.
| Usage | Current rate | Approximate coverage from $200 |
|---|---|---|
| Nova-3 Mono pre-recorded | $0.0043/min | 46,512 minutes, about 775 hours |
| Nova-3 Mono streaming | $0.0048/min | 41,667 minutes, about 694 hours |
| Nova-3 Multi streaming | $0.0058/min | 34,483 minutes, about 575 hours |
| Flux English streaming | $0.0065/min | 30,769 minutes, about 513 hours |
Those estimates exclude paid add-ons. That is actually useful for testing: instead of spending the credit on clean sample clips, run the audio you expect in production. Include long recordings, short recordings, multiple speakers, specialist terms, poor microphones and the features your finished product will enable.
Pay As You Go vs Growth: The discount only works if you consume the credits
| Plan | Starting commitment | Pricing structure | Useful for |
|---|---|---|---|
| Pay As You Go | $200 free credit, then usage | No annual minimum | Testing, variable workloads and smaller deployments |
| Growth | $4,000+ per year | Prepaid annual credits with discounted rates | Predictable production usage across Deepgram products |
| Enterprise | Custom | Negotiated | Higher volume, deployment, support or data requirements |
Growth can reduce unit prices by up to 20% depending on the product, but the headline discount is not the only factor in the calculation. The annual commitment starts at $4,000 in prepaid credits, so your expected consumption must justify it.
For example, 10,000 hours a year of Nova-3 Monolingual pre-recorded transcription would cost about $2,160 at the $0.0036 Growth unit rate. If that were your only Deepgram workload, it would not, by itself, consume a $4,000 annual prepay. A team also buying streaming, TTS, or voice-agent minutes can reach the threshold much sooner.
Growth can also make sense because of capacity rather than price. Pay As You Go currently allows up to 150 simultaneous speech-to-text WebSocket connections, while Growth raises that published limit to 225. For a live product, peak concurrency can force the commercial decision before total monthly minutes do.
Hidden Deepgram pricing variables that can change the invoice
The basic arithmetic is easy. These are the parts worth putting into a production spreadsheet before launch.
| Variable | Why does it change cost | What to model |
|---|---|---|
| Temporary streaming prices | Several current rates are promotional rather than permanent list prices. | Run both current-rate and regular-rate scenarios. |
| Audio channels | Deepgram calculates multichannel billing using the processed duration across channels. A 10-minute two-channel file can therefore represent 20 billable channel-minutes. | Use channel-minutes, not just file duration. |
| Add-ons | Redaction, keyterm prompting, entity detection and streaming diarisation stack on top of STT. | Price the finished transcript specification. |
| Billing granularity | Deepgram states that it uses true per-second billing rather than rounding each request up to a full minute. | This matters most with large volumes of short clips. |
| Retries and duplicate jobs | Application-level retry logic can resend the same audio if failures are handled poorly. | Track submitted audio against successful unique jobs. |
| Concurrency | Peak live sessions can hit plan limits even when monthly usage is modest. | Forecast concurrent sessions separately from monthly minutes. |
| Credit auto-load | Deepgram says Pay As You Go accounts automatically reload $100 when the remaining credit falls below $10. | Review auto-load settings before load tests or public launches. |
| Downstream processing | Summaries, LLM calls, storage, search and other processing may sit outside the STT line item. | Calculate the cost per finished workflow, not just transcription. |
See more of our reporting in Google Top Stories, AI Overviews and AI Mode.
Deepgram text-to-speech pricing has a temporary Flux TTS promotion
Text-to-speech uses a different unit of measure: characters rather than audio minutes. Deepgram’s new Flux TTS is currently free through 12 September 2026. Standard pricing begins on 13 September 2026.
| TTS model | Pay As You Go | Growth |
|---|---|---|
| Flux TTS | Free through 12 September 2026, then $0.0450/1,000 characters | Free through 12 September 2026, then $0.0405/1,000 characters |
| Aura-2 | $0.030/1,000 characters | $0.027/1,000 characters |
| Aura-1 | $0.0150/1,000 characters | $0.0135/1,000 characters |
Do not convert those character rates directly into a fixed cost per spoken minute. Speech duration depends on the words, punctuation and delivery. For a production voice application, estimate the characters generated per conversation from real transcripts and then apply the character rate.
Deepgram Voice Agent API pricing changes on 13 September 2026
Voice Agent API pricing is materially higher than raw STT because you are buying an agent session rather than a transcript. Deepgram bills it per minute of connection time, with different rates depending on which parts of the stack you bring yourself.
Flux TTS is free inside the Voice Agent API through 12 September 2026. This temporarily reduces several tiers to their BYO TTS rate.
| Voice Agent tier | PAYG through 12 Sep | PAYG from 13 Sep | Growth through 12 Sep | Growth from 13 Sep |
|---|---|---|---|---|
| Standard | $0.056/min | $0.075/min | $0.051/min | $0.068/min |
| Standard – BYO TTS | $0.065/min | $0.065/min | $0.051/min | $0.051/min |
| Custom – BYO LLM | $0.050/min | $0.065/min | $0.041/min | $0.059/min |
| Custom – BYO LLM + TTS | $0.050/min | $0.050/min | $0.041/min | $0.041/min |
| Advanced | $0.122/min | $0.163/min | $0.110/min | $0.146/min |
| Advanced – BYO TTS | $0.122/min | $0.122/min | $0.110/min | $0.110/min |
A product using 50,000 Standard Voice Agent minutes in August 2026 would have a Deepgram line item of about $2,800 at the temporary rate of $0.056 per minute. At the $0.075 rate beginning 13 September, the same 50,000 minutes would cost $3,750.
BYO tiers should not be described as automatically cheaper. They reduce what Deepgram charges for the relevant component, but the LLM or TTS provider you bring still has its own bill. The correct comparison is total cost per completed agent conversation across all providers.
Audio Intelligence uses tokens rather than audio minutes
Deepgram’s Audio Intelligence features use another billing unit. Summarisation, topic detection, sentiment analysis and intent recognition are currently priced at $0.0003 per 1,000 input tokens and $0.0006 per 1,000 output tokens on Pay As You Go. Growth rates are $0.00024 and $0.00048, respectively.
This cost can be small beside the audio pipeline, but it should still sit on a separate line in your model. Mixing STT minutes, TTS characters, Voice Agent connection minutes, and Audio Intelligence tokens into a single assumed “Deepgram minute” makes forecasts unreliable.
Deepgram vs Whisper pricing: compare the workflow, not one rate
If your requirement is simple file transcription, Deepgram is not the only low-cost route. OpenAI now has several managed transcription models as well as the older Whisper API, and self-hosted Whisper changes the cost model again. Our OpenAI Whisper API pricing guide converts those options into per-minute, hourly and high-volume costs.
The useful distinction for budgeting is architectural. Pre-recorded transcription lets you compare the cost per processed audio hour directly and fairly. Real-time speech is harder because model latency, speaker features, concurrency, session handling, and the surrounding agent stack affect the effective cost.
Deepgram pricing FAQs
How much does Deepgram speech-to-text cost per minute in 2026?
Nova-3 Monolingual currently costs $0.0043 per minute for pre-recorded transcription and a promotional $0.0048 per minute for streaming on Pay As You Go. Nova-3 Multilingual costs $0.0052 per minute pre-recorded and a promotional $0.0058 per minute for streaming.
How much does Deepgram cost per audio hour?
Nova-3 Monolingual pre-recorded costs about $0.258 per hour, usually rounded to $0.26 for budgeting. Its current streaming promotional rate comes to $0.288 per hour, or about $0.29.
How much does Deepgram Nova-3 cost?
On Pay As You Go, Nova-3 Monolingual is $0.0043 per minute for pre-recorded content and $0.0048 per minute for streaming. Nova-3 Multilingual is $0.0052 per minute for pre-recorded content and $0.0058 per minute for streaming. Growth has lower rates in each of those Nova-3 routes.
Does Deepgram have a free plan or free credits?
Deepgram currently gives new Pay-As-You-Go accounts $200 in free credit and does not require a credit card to start. At the current Nova-3 Monolingual streaming rate, $200 represents roughly 694 hours before paid add-ons.
Is Deepgram streaming more expensive than pre-recorded transcription?
For Nova-3, yes. Current Pay As You Go pricing is $0.0043 per minute for pre-recorded Nova-3 Monolingual, versus a promotional $0.0048 per minute for streaming. The larger practical difference is that pre-recorded speaker diarisation is included, while streaming diarisation costs $0.0020 per minute.
Does Deepgram charge extra for speaker diarisation?
It depends on the mode. Speaker diarisation is currently included for pre-recorded speech-to-text. For streaming, it costs $0.0020 per minute on Pay As You Go and $0.0017 per minute on Growth.
How does Deepgram bill multichannel audio?
Deepgram calculates billing from the total processed audio across channels. A 10-minute file containing two separately processed audio channels can therefore represent 20 billable channel-minutes. This is easy to miss in call-recording systems where customer and agent audio sit on separate channels.
Does Deepgram round audio up to the nearest minute?
No. Deepgram states that it uses true per-second billing. That makes a bigger difference for workloads with thousands of short clips than for long meetings or calls.
How much does the Deepgram Voice Agent API cost?
As of 24 August 2026, the Standard Voice Agent API costs $0.056 per minute on Pay As You Go because Flux TTS is temporarily free inside the agent stack. The Standard rate changes to $0.075 per minute on 13 September 2026. BYO and Advanced tiers use different rates.
When should you consider Deepgram Growth pricing?
Growth is worth modelling once your annual Deepgram usage is predictable enough to consume at least the prepaid commitment, which starts at $4,000 per year. Higher concurrency requirements can also make Growth relevant even before unit-price savings become the main reason to upgrade.
How to budget for Deepgram without underestimating the bill
Start with five numbers: audio minutes, processing mode, model, required add-ons and audio channel count. That gives you a realistic speech-to-text baseline. For live applications, add expected peak concurrency and retries. For voice agents, calculate the whole session rather than substituting the much cheaper Nova-3 STT rate.
For recorded single-language audio, the current starting point is Nova-3 Monolingual pre-recorded at about $0.26 per hour. For live Nova-3 transcription, it is about $0.29 per hour under the current promotion. From there, add the features your output actually requires.
That keeps this page’s job simple: work out what Deepgram will cost. Once the numbers fit your budget, use the separate review to decide whether its accuracy, latency and workflow are good enough for the application you are building.
You Might Also Like:
Best AI Speech To Text Tools 2026
OpenAI Whisper API Pricing
OpenAI Whisper Review 2026
