Deepgram is a STT / ASR company founded in 2015 and based in San Francisco, United States. It has raised $214M in total funding, most recently a Series C in 2026 at a $1.3B valuation.
| Date | Stage | Amount | Valuation | Lead investors |
|---|---|---|---|---|
| Jan 13, 2026 | Series C | $130M | $1.3B | AVP |
| Nov 29, 2022 | Series B | $47M | — |
| Madrona |
| Feb 3, 2021 | Series B | $25M | — | Tiger Global |
| Mar 1, 2020 | Series A | $12M | — | Wing VC |

Deepgram launches real-time on-device voice AI for Snapdragon-based PCs.
Deepgram discusses voice AI infrastructure and production-grade agent development in a podcast interview.
Deepgram discusses voice AI infrastructure and production-grade agent deployment in a podcast interview.
Deepgram's latest STT model delivering industry-leading accuracy with a 54.2% reduction in word error rate for streaming versus prior benchmarks. Supports real-time multilingual conversation transcription, domain-specific terminology enhancement, and optional PII redaction. Available for both streaming and batch workloads; used by Spotify, NASA, and Twilio for high-volume production speech-recognition pipelines.
An enterprise-grade TTS engine delivering sub-200ms streaming latency, optimized for voice agent applications requiring low cost at thousands of concurrent sessions. Recognized with the 2025 Contact Center Technology Award. Designed for on-prem and cloud deployment with domain-specific pronunciation accuracy. Winner of the vertically integrated pricing model at $4.50/hour when paired with Deepgram's full STT stack.
A unified speech-to-speech API (GA June 2025) that combines Nova-3 STT, LLM orchestration, and Aura-2 TTS into a single real-time interface for building context-aware voice agents. Supports bring-your-own LLM and TTS configurations. Gives developers full control over conversational state, turn detection, and interruption handling. Targets enterprise voice-agent developers who need production-grade reliability across thousands of concurrent sessions.
3 patents on file, but none with both an extractable figure and an abstract on Google Patents yet.