Deepgram review

4.1/5

Voice AI API platform (STT/TTS/Voice Agent)

Deepgram is a developer-first voice AI platform offering speech-to-text (Nova-3, Flux), text-to-speech (Aura), and a Voice Agent API behind a single usage-based API. It competes on low latency (sub-250ms), high transcription accuracy, and aggressive per-minute pricing, and is built for engineers wiring real-time and batch audio into their own products rather than for non-technical buyers.

Visit Deepgram → last verified 2026-06-02

Affiliate link · how we make money

Best for voice AI APIs. Deepgram starts Free $200 credit, then usage-based ~$0.0048/min (Nova-3 STT) on a Usage-based (per-minute STT / per-1k-char TTS / per-minute Voice Agent), with $200 free credit and an optional prepaid annual Growth tier model. Transcribing 100 hours/mo of audio with Nova-3 streaming at $0.0048/min runs 100 x 60 x $0.0048 = $28.80/mo. The same 100 hours pre-recorded at $0.0077/min (~$0.46/hr) is about $46/mo. A 24/7 Standard Voice Agent at $0.075/min for ~10,000 conversation minutes/mo is about $750/mo.

Spec

pricing modelUsage-based (per-minute STT / per-1k-char TTS / per-minute Voice Agent), with $200 free credit and an optional prepaid annual Growth tier
entry priceFree $200 credit, then usage-based ~$0.0048/min (Nova-3 STT)verify livedecode →
free tier$200 free credit on Pay As You Go, no credit card required, no expirationsource →
self-hostYes
MCPServer
integrationsSDKs (Python, JS/Node, .NET, Go, Rust), official MCP server (dg CLI) for Claude Code/Cursor/Windsurf, self-hosted via Docker/Kubernetes/SageMakersource →
API accessYes
affiliateAffiliate/partner program runs via PartnerStack; specific commission rate not publicly disclosed (also listed on FlexOffers)

The gotcha

the pricing catch

Headline per-minute rates are for the base Nova-3 model only; add-ons like diarization (+$0.0020/min), redaction (+$0.0020/min), and keyterm prompting (+$0.0013/min) stack on top, and multilingual/Flux models cost more, so a fully featured stream can cost well above the advertised $0.0048/min. The ~20% Growth discount also requires committing $4K+/year in prepaid credits.

Decode the real cost →

Best for

  • You are a developer building voice features (transcription, TTS, or voice agents) and want a fast, cheap, usage-billed API with optional self-hosting, rather than a no-code product.

source: deepgram.com/pricing →

Deepgram pricing tiers

Prices move. Cells flagged verify link to the live vendor page.

Decode Deepgram pricing →
tiermonthlyannualincludedunit
Pay As You Go (STT - Nova-3) Usage-basedverify n/a $200 free credit, no card; Nova-3 streaming $0.0048/min, pre-recorded $0.0077/min; multilingual $0.0058/$0.0092 per minute
Pay As You Go (TTS - Aura) Usage-basedverify n/a Aura-2 $0.030 / 1k characters; Aura-1 $0.0150 / 1k characters per 1k characters
Pay As You Go (Voice Agent API) Usage-basedverify n/a Standard $0.075/min, BYO-TTS $0.065/min, BYO-LLM $0.056/min, Advanced $0.163/min per minute
Growth ~$333 ($4K+/year prepaid)verify n/a Up to ~20% off PAYG rates via prepaid annual credits, higher concurrency limits prepaid annual credits
Enterprise Customverify n/a Volume pricing, self-hosted/on-prem deployment, dedicated support, SLAs custom (contact sales)

// pools and per-unit rates are volatile · cells flagged verify link to the live vendor page

Compare Deepgram with…

Questions

How much does Deepgram cost?

Deepgram starts Free $200 credit, then usage-based from ~$0.0048/min (Nova-3 STT). Transcribing 100 hours/mo of audio with Nova-3 streaming at $0.0048/min runs 100 x 60 x $0.0048 = $28.80/mo. The same 100 hours pre-recorded at $0.0077/min (~$0.46/hr) is about $46/mo. A 24/7 Standard Voice Agent at $0.075/min for ~10,000 conversation minutes/mo is about $750/mo.

Can you self-host Deepgram?

Yes, Deepgram can be self-hosted.

Does Deepgram support MCP?

Deepgram supports MCP as Server, so it works with assistants like Claude and ChatGPT.

What's the catch with Deepgram's pricing?

Headline per-minute rates are for the base Nova-3 model only; add-ons like diarization (+$0.0020/min), redaction (+$0.0020/min), and keyterm prompting (+$0.0013/min) stack on top, and multilingual/Flux models cost more, so a fully featured stream can cost well above the advertised $0.0048/min. The ~20% Growth discount also requires committing $4K+/year in prepaid credits.