Introduction
Deepgram sells speech-to-text (STT), text-to-speech (TTS) and a hosted Voice Agent API, and nearly all of it is billed by usage. The public pricing page has three plans: Pay As You Go, Growth and Enterprise. New accounts get $200 in free credit, and no card is needed to claim it. The base plans are easy to follow. The hard part is working out what you will actually pay once add-ons, promotional rates and concurrency limits come into play.
This guide uses Deepgram's own pricing page as it stood in October 2026. scored.tools gives Deepgram an overall rating of 7.4/10. The criteria scores are 9 for ease of use, 9 for API and integrations, 8 for pricing value, 6 for problem fit and 5 for output quality. Output quality scores low because the only accuracy and latency claims available come from Deepgram itself, with no independent benchmark behind them.
Pricing Tiers Table
| Plan | Price | What you commit | Main extras |
|---|---|---|---|
| Pay As You Go (with free credit) | Per-use rates, $200 free credit for new accounts | Nothing. No minimums, no expiry | All models, API keys, playground, community support |
| Growth | From $4,000/year in prepaid credits | Annual prepay | Up to 20% lower rates (Deepgram's figure), higher streaming concurrency |
| Enterprise | Not published | Sales contract | Custom volume pricing, custom concurrency, self-hosting, BAA |
Speech-to-text rates
| Model | Mode | PAYG per min | Growth per min | PAYG per hour |
|---|---|---|---|---|
| Nova-3 Monolingual | Pre-recorded | $0.0043 | $0.0036 | $0.26 |
| Nova-3 Multilingual | Pre-recorded | $0.0052 | $0.0043 | $0.31 |
| Whisper Large | Pre-recorded | $0.0048 | $0.0048 | $0.29 |
| Nova-3 Monolingual | Streaming | $0.0048 (promo, regular $0.0077) | $0.0042 (regular $0.0065) | $0.29 |
| Nova-3 Multilingual | Streaming | $0.0058 (promo, regular $0.0092) | $0.0050 (regular $0.0078) | $0.35 |
| Flux English | Streaming | $0.0065 (promo, regular $0.0077) | $0.0057 (regular $0.0065) | $0.39 |
| Flux Multilingual | Streaming | $0.0078 | $0.0068 | $0.47 |
Text-to-speech rates (per 1,000 characters)
| Model | PAYG | Growth |
|---|---|---|
| Aura-1 | $0.0150 | $0.0135 |
| Aura-2 | $0.030 | $0.027 |
| Flux TTS | $0.0450 | $0.0405 |
Voice Agent API rates
| Tier | PAYG per min | Growth per min |
|---|---|---|
| Standard | $0.075 | $0.068 |
| Standard, bring your own TTS | $0.065 | $0.051 |
| Custom, bring your own LLM | $0.065 | $0.059 |
| Custom, bring your own LLM and TTS | $0.050 | $0.041 |
| Advanced | $0.163 | $0.146 |
| Advanced, bring your own TTS | $0.122 | $0.110 |
What Each Tier Gets You
Pay As You Go and the $200 credit
Deepgram doesn't have a permanent free plan. What it has is $200 of credit on a new Pay As You Go account. Deepgram says this covers roughly 43,000 minutes of Nova transcription, and the page states that the credit doesn't expire. That is more than most rivals give: AssemblyAI gives $50, Retell AI gives $10 and Vapi gives $5. For an indie developer, $200 is enough to build a prototype and run a pilot on real audio before any money changes hands.
Once the credit runs out, you pay list rates with no minimum. Every model is available, along with SDKs for JavaScript, Python, .NET, Go and Java, plus integrations with Twilio, Amazon Connect, Zoom, LiveKit and Zapier. Concurrency limits are 50 REST STT requests, 150 streaming STT connections, 45 TTS connections and 45 Voice Agent connections. Deepgram's hosted Whisper is capped at 5 concurrent requests.
Growth
Growth starts at $4,000 a year in prepaid credit. Deepgram says it saves up to 20%. On the main models, the discount works out at around 12 to 17% off PAYG. Nova-3 pre-recorded falls from $0.0043 to $0.0036 a minute, and the Voice Agent Standard tier falls from $0.075 to $0.068. Streaming STT concurrency goes up from 150 to 225, and TTS and Voice Agent limits go up from 45 to 60. If you spend more than your prepaid credit, the overage is billed at the Growth rate plus 10%.
Enterprise
Enterprise is the only plan that offers self-hosted deployment, custom model training, custom concurrency and a BAA for HIPAA work. None of these terms are published, so you will need to talk to sales. Few voice AI vendors let you self-host, which makes this the real reason regulated teams pick Deepgram. If you only need cheaper rates, Growth is probably enough.
Hidden Costs
Add-ons stack per minute. Redaction costs $0.0020 a minute on PAYG, entity detection $0.0017 and keyterm prompting $0.0013. Speaker diarization is free on pre-recorded audio, but on streaming it adds $0.0020 a minute. Smart formatting is free everywhere. A streaming Nova-3 call with diarization and redaction comes to $0.0088 a minute on PAYG ($0.0048 + $0.0020 + $0.0020), about 83% more than the headline rate. At 10,000 minutes a month, that is $88 instead of $48.
Several streaming rates are promotional. The pricing page shows struck-through regular prices next to the Nova-3 and Flux English streaming rates. Nova-3 Monolingual streaming normally costs $0.0077 a minute, against the current $0.0048. Deepgram doesn't say when the promotion ends, so build your budget around the regular rate and treat the gap as a temporary saving.
Multilingual costs more. On pre-recorded audio, Nova-3 Multilingual costs about 21% more than the monolingual model. Flux Multilingual streaming costs $0.0078 against $0.0065 for Flux English.
Voice Agent billing counts connection time. Deepgram bills the agent API on WebSocket connection time, so a call left open in silence still costs money. Telephony isn't on the pricing page either. Twilio or your SIP carrier will charge you separately.
Legacy models cost more. According to the FAQ, Enhanced is $0.99 an hour and Base is $0.87 an hour on PAYG, both well above Nova-3. If your code still points at an old model, switching to Nova-3 is the quickest saving available.
TTS rates vary by 3x. The tool data quotes $0.0150 per 1,000 characters, which is Aura-1. Aura-2 costs twice that and Flux TTS three times. Until December 31, 2026, Deepgram will match Flux TTS purchases with credit up to $500, which cuts the cost of an early trial.
How It Compares to Competitors
| Vendor | Batch STT per hour | Streaming STT per hour | TTS per 1K chars | Agent per min | Free credit |
|---|---|---|---|---|---|
| Deepgram | $0.26 (Nova-3) | $0.29 promo / $0.46 regular (Nova-3) | $0.015 to $0.045 | $0.075 (Standard) | $200 |
| AssemblyAI | $0.15 (Universal-2), $0.21 (Universal-3.5 Pro) | $0.15 (Universal-Streaming), $0.45 (U3.6 Pro Realtime) | n/a | n/a | $50 |
| ElevenLabs | $0.22 (Scribe v2) | $0.39 (Scribe v2 Realtime) | $0.04 (Flash/Turbo), $0.08 (Multilingual) | $0.08 | Not listed on API page |
| Vapi | n/a | Passed through at cost | Passed through at cost | $0.05 platform fee plus model costs | $5 |
| Retell AI | n/a | Bundled | From $0.015/min | About $0.11 in Retell's own example | $10 |
On raw transcription price, Deepgram loses. AssemblyAI's Universal-2 batch model costs $0.15 an hour, against $0.26 for Nova-3, and ElevenLabs' Scribe v2 costs $0.22. In streaming, AssemblyAI's Universal-Streaming also costs $0.15 an hour. Deepgram's promotional Nova-3 rate is close to Scribe v2 Realtime but more than AssemblyAI's base streaming tier. AssemblyAI also bills streaming on session time, and its add-ons stack in the same way, so compare full configurations rather than headline rates.
Deepgram's best price case is in TTS and agents. Aura-2 costs $0.030 per 1,000 characters, compared with $0.04 for ElevenLabs Flash and $0.08 for its Multilingual voices. The tool data still rates ElevenLabs ahead on expressive voices and cloning, so cheaper TTS only helps if the Aura voices are good enough for your product. On agents, Deepgram's Standard tier at $0.075 a minute bundles STT, TTS and LLM orchestration. ElevenLabs charges $0.08 a minute. Retell's own estimator example comes to $0.11 a minute. Vapi's $0.05 fee looks lower, but it is only the platform fee: transcription, LLM and voice are added on top at cost.
Which Plan Should You Pick
Building a prototype or MVP
Use Pay As You Go and spend the $200 credit. You don't commit to anything, and at Nova-3 batch rates the credit lasts a long time. Try your real add-on mix (diarization, redaction, keyterms) during the trial so the per-minute figure you budget with is accurate.
Spending $400 or more a month on Deepgram
Look at Growth. At a 12.5% discount, $4,000 of prepaid credit buys about $4,570 of usage at PAYG rates, a saving of roughly $570 a year. If your monthly spend is unpredictable and might drop, the prepay is a risk. Below about $330 a month, stay on PAYG.
Running a voice agent product
If you already have an LLM contract or a preferred TTS vendor, the bring-your-own tiers are the ones to price. Custom with your own LLM and TTS costs $0.050 a minute on PAYG and $0.041 on Growth. The Advanced tier, at $0.163 a minute, is more than double Standard, so check whether you need it before choosing it.
Healthcare, finance or government
Go to Enterprise. Only Enterprise offers self-hosting and a BAA, and you will be negotiating volume pricing at that level anyway.
Verdict
Deepgram has clear, published per-minute pricing, the largest free credit in this group and a self-hosting option that few competitors match. Those are the reasons for its 8/10 pricing value score and 7.4 overall. It is not the cheapest place to transcribe files. AssemblyAI and ElevenLabs both charge less per batch hour, and the gladia.io review linked in the tool data notes that Deepgram is weaker for occasional file transcription than for real-time voice. Choose Deepgram if you are building streaming or agent products and value bundled pricing, high concurrency and a self-hosted option. Before committing, budget at the regular streaming rates rather than the promotional ones, and include your add-ons in the per-minute figure.
