Deepgram is a developer-focused voice AI API covering speech-to-text transcription, text-to-speech (Aura), and a combined Voice Agent API for building real-time conversational products. It's built for engineering teams shipping production features — transcription for call centers and meetings, voice bots for support and sales, or narration for apps — rather than for solo creators editing one file. Pricing is entirely usage-based: pay-as-you-go rates per minute or per 1,000 characters, a prepaid "Growth" tier at a discount, custom Enterprise deals, and $200 in free credit to start.
Who builds it
Deepgram was founded in 2015 by CEO Scott Stephenson and a co-founder who met doing machine learning research (including waveform analysis for a dark matter detector) before turning to speech technology. The company is remote-first, with staff across 20+ U.S. states and 5+ countries, and describes itself as a research-driven, foundational AI company focused on voice technology. In January 2026 it raised a $130 million Series C, pushing its valuation to $1.3 billion.
Core features
- Nova-3 — production speech-to-text model supporting 50+ languages, with monolingual and multilingual pricing tiers
- Flux — real-time conversational speech recognition built for voice agents, with built-in turn detection; currently covers English, Spanish, German, French, Hindi, Russian, Portuguese, Japanese, Italian and Dutch
- Aura-2 and Aura-1 — text-to-speech models with 40+ English voices spanning US, Australian and Philippine accents, built for sub-200ms streaming in voice-agent use cases
- Voice Agent API — bundles STT, an LLM, and TTS into one low-latency pipeline, billed on websocket connection time
- Add-ons: speaker diarization and PII/text redaction, priced per minute on top of base transcription
- Industry-tuned model variants for healthcare, legal and finance vocabulary, plus public cloud, private cloud and on-premises deployment options
Pricing
- Speech-to-text (pay-as-you-go, per minute): Flux English $0.0065–$0.0077; Nova-3 Monolingual $0.0048–$0.0077; Nova-3 Multilingual $0.0058–$0.0092; diarization/redaction add-ons +$0.0020/minute
- Text-to-speech (per 1,000 characters): Aura-2 $0.030 pay-as-you-go / $0.027 on Growth; Aura-1 $0.015 pay-as-you-go / $0.0135 on Growth
- Voice Agent API: roughly $0.050–$0.163/minute depending on tier and configuration
- Growth plan: prepaid annual credits at up to ~20% off pay-as-you-go rates (about 13% lower on most STT models)
- Enterprise: custom volume pricing and terms
- New accounts get $200 in free credit, no credit card required to start; check the current pricing page as rates can change
Who it's for
Deepgram fits engineering teams building transcription, voice agents, or TTS narration directly into a product, where usage-based per-minute pricing and low-latency streaming matter more than a point-and-click editor. The $200 free credit and pay-as-you-go tier suit prototyping and small apps; the Growth and Enterprise tiers make sense once volume is predictable enough to prepay for the discount, or once a team needs private-cloud/on-prem deployment and custom contract terms.