Deepgram Pricing: Plans, Costs & What You Get (2026)

Deepgram's speech-to-text starts at $0.0043 per minute for pre-recorded Nova-3, but that is the floor of one product line, but that's just the start. Text-to-speech is billed per character, the Voice Agent API blends both into one meter, and a matching-credit promotion is currently running on Flux TTS.

I went through Deepgram's entire rate card, plan by plan and product by product, to see where the real bill comes from once you stack all four products together.

This guide breaks down what each product costs today and how the pieces add up once you're running a real agent.

TL;DR: Deepgram Pricing

• Deepgram's cheapest speech-to-text rate is $0.0043 per minute for pre-recorded Nova-3, while streaming starts at $0.0048 per minute, a limited-time promotional rate against a regular price of $0.0077 per minute.
• Text-to-speech is billed per character. Aura-2 costs $0.030 per 1,000 characters.
• Flux TTS costs $0.045 per 1,000 characters on Pay As You Go and $0.0405 on Growth. Deepgram is currently matching Flux TTS spend with credits up to $500, through December 31, 2026.
• The Voice Agent API bundles STT, LLM routing, and TTS into one per-minute meter, and the Standard and Advanced tiers include Deepgram's own TTS in that rate.
• Growth plans save up to 20% against Pay As You Go rates but need a $4,000-a-year prepaid commitment.

Deepgram Pricing at a Glance

Pricing verified against deepgram.com/pricing on September 4, 2026. Verify with Deepgram before you budget.

Deepgram Pricing Plans Breakdown

Pay As You Go: Free $200 Credit, Then Usage-Based

What's included: Full access to every Deepgram product at standard rates, with no minimum spend and no credit expiry. Pre-recorded STT starts at $0.0043/min, Aura-2 TTS runs $0.030/1k characters, and the Voice Agent API runs $0.050 to $0.163/min.

A $200 credit covers testing, with support through Deepgram's community channels.

Best for: Developers and startups testing before committing.

Pros

✅ No commitment and no minimum spend, so you only pay for what you actually use

✅ The $200 free credit covers a genuine test, not just a demo call

Cons

❌ Every product runs at its highest per-unit rate on this plan

❌ Lower concurrency ceiling than Growth, which matters once you're running more than a handful of calls at once

Growth: $4,000+ Per Year Prepaid

What's included: The same four products at up to 20% off, in exchange for a prepaid annual commitment. Aura-2 TTS drops to $0.027/1k characters, and the Voice Agent API's Custom BYO LLM + TTS tier drops to $0.041/min. Concurrency rises to 225 WebSocket streams for STT, up from 150 on Pay As You Go.

Best for: Applications with predictable, growing volume that can commit to annual spend.

Pros

✅ Up to 20% off every product line, which makes a big difference at higher volumes

✅ Meaningfully higher concurrency for streaming STT, which is where Pay As You Go limits get hit first

Cons

❌ The $4,000-a-year minimum applies even if actual usage comes in lower

❌ Support doesn't upgrade from Pay As You Go despite the added commitment

❌ Overages are billed at the Growth rate plus a 10% premium plus tax, calculated every Monday for the previous week. A valid credit card must stay on file, and API requests return a 402 error once credits are exhausted without one.

Enterprise: Custom Quote

What's included: Custom volume discounts beyond Growth's 20%, dedicated support, custom SLAs, and deployment options including self-hosted and private-cloud instances not available on the self-serve plans.

Best for: Large-volume, custom deployment, or dedicated support needs.

Pros

✅ Deepest available discounts and the highest concurrency ceilings

✅ Deployment flexibility, including self-hosted and private-cloud, that no self-serve plan offers

Cons

❌ No self-serve signup. Getting a quote means a sales conversation

❌ Pricing isn't public, so you can't budget precisely until you've talked to Deepgram

Which Deepgram Pricing Path Should You Choose?

Choose Pay As You Go if you:

• Are still testing which model or product fits your agent and don't want a commitment yet.
• Have low or unpredictable monthly volume.

Choose Growth if you:

• Can commit to $4,000+ in annual usage and want the 20% discount locked in.
• Need higher concurrency limits (225 WSS connections for streaming STT versus 150 on Pay As You Go).

Choose Enterprise if you:

• Need custom deployment, like self-hosted or private-cloud instances.
• Require dedicated support or contract terms that self-serve plans don't cover.

Is Deepgram Worth the Cost?

Deepgram's rate card is transparent, which is rare in this category. But five metered surfaces, two buying paths, and the September 15 reprice on the Flux TTS tiers mean the real cost requires some calculating beyond a glance at the homepage number.

Deepgram is worth it if you:

• Need low-latency streaming transcription and are comparing it against per-minute rates from Google, AWS, or Azure.
• Want one vendor covering STT, TTS, and voice-agent orchestration instead of assembling three separately.

Skip Deepgram if you:

• Need the absolute lowest transcription rate. AssemblyAI's Universal-2 runs $0.15 per hour, about $0.0025 per minute, and OpenAI's gpt-4o-mini-transcribe runs $0.003 per minute. Both undercut Deepgram's $0.0043 pre-recorded rate.
• Are building on your own STT/TTS stack already and only need turn-taking logic, not a full managed agent layer.

Deepgram Alternatives & Pricing Comparison

These four cover the same transcription job at very different rates. Read the starting prices as floors: each vendor meters streaming, add-ons, and agent orchestration separately.

Cekura vs. Deepgram: Which Should You Choose?

Cekura is better for:

• Catching failures before a caller hits them, through pre-production simulation across interruptions, background noise, and adversarial scenarios.
• Watching live call quality once a Deepgram-based agent is in production, including drop-off points and compliance check failures.

Deepgram is better for:

• Converting speech to text, generating spoken replies, or running managed voice-agent orchestration.
• Companies that want STT, TTS, and LLM routing under one vendor and one bill.

Use both if:

• You're running a Deepgram-powered agent in a regulated setting like healthcare, where a missed branch in a call flow means a real appointment gets missed.
• You can only manually check a small slice of production calls and need automated coverage for the rest.

Cekura's side of that covers three stages of a voice agent's life:

• Pre-production: Simulate thousands of caller scenarios before a Deepgram-powered agent goes live.
• Infrastructure: Catch latency and turn-taking issues under concurrent load, the same layer where per-minute costs and containment rates both move.
• Observability: Monitor live calls for instruction-following, sentiment, and drop-off points once the agent is in production.

Native integrations work out of the box for Retell, Vapi, ElevenLabs, LiveKit, Pipecat, Bland AI, Agora, Genesys, and Kore.ai**.** You don't rebuild anything. You add a testing and monitoring layer on top of what you already have, including a Deepgram-based STT pipeline.

Cekura supports SOC 2, HIPAA, and GDPR compliance, covering transcript redaction, role-based access, and audit trails.

Per Cekura's benchmarks, task completion and latency vary widely across orchestration stacks, so the cost of a slower pipeline shows up on Deepgram's per-minute meter as well as in customer experience. See https://benchmarks.cekura.ai/ for the current numbers.

Try Cekura for free here.

The Bottom Line on Deepgram Pricing

Deepgram's rate card is transparent, which puts it ahead of vendors that gate pricing behind a demo call. The complicated part is that pricing spreads across four product lines and two buying paths.

Budget against the rates on the card today: $0.045 per 1,000 Flux TTS characters, $0.075 per minute on Standard, and $0.163 per minute on Advanced. Streaming speech-to-text is on promotional pricing, so budget against the $0.0077 regular rate rather than the $0.0048 headline.

A concrete case: 10,000 agent minutes a month on the Standard Voice Agent tier costs $750.

The same volume on Custom BYO LLM and TTS costs $500, plus whatever your own LLM and TTS providers charge. On Growth, those become $680 and $410, against a $4,000 annual prepayment that 10,000 minutes a month would clear in under six months.

Model your own average handle time before you compare tiers. It moves the bill more than any single pricing decision on this page.

Frequently Asked Questions

How much does Deepgram cost per minute?

Deepgram's speech-to-text starts at $0.0043 per minute for pre-recorded Nova-3 audio and $0.0048 per minute for streaming. Text-to-speech is billed per character instead, and the Voice Agent API runs $0.050 to $0.163 per minute depending on tier and whether you bring your own LLM or TTS.

Is Deepgram Flux TTS really free?

Flux TTS is not free. It costs $0.045 per 1,000 characters on Pay As You Go and $0.0405 on Growth. Deepgram is currently matching Flux TTS spend with credits up to $500 on both plans, through December 31, 2026.

What is the difference between Deepgram Pay As You Go and Growth?

The main difference between Pay As You Go and Growth is commitment and discount. Pay As You Go has no minimum and starts with a $200 free credit, while Growth requires a $4,000-a-year prepaid commitment in exchange for up to 20% off most rates.

Does the Voice Agent API include LLM and TTS costs?

Yes, the Voice Agent API's Standard and Advanced tiers bundle STT, LLM routing, and TTS into one per-minute rate. The Custom-BYO tiers let you swap in your own LLM or TTS provider and pay Deepgram only for orchestration.

Is Deepgram cheaper than AssemblyAI or OpenAI?

AssemblyAI is cheaper on both. Universal-2 async and Universal-Streaming both run $0.15 per hour, about $0.0025 per minute, against Deepgram's $0.0043 pre-recorded and $0.0048 promotional streaming rate.

Deepgram is cheaper than OpenAI's live transcription, which runs $0.017 per minute. For AssemblyAI's full rate card, see our AssemblyAI pricing guide. The right pick depends on whether you need real-time or batch, and on how many add-on meters you switch on.