Deepgram

Deepgram is a foundational voice AI company providing developer-first APIs for speech-to-text transcription, text-to-speech synthesis, and real-time conversational voice agents. The company was founded on August 18, 2015 by Scott Stephenson, Adam Sypniewski, and Noah Shutty, three University of Michigan physicists who had previously worked on dark matter detection research at a lab in China. The origin story is genuinely unusual: Stephenson and Shutty had built a wearable device to record audio around the clock as a personal life-logging experiment, but found no good way to search or index the hundreds of hours of audio they accumulated. Realizing the waveform-analysis techniques from their physics research could apply to speech, they built what would become Deepgram, going through Y Combinator's Winter 2016 batch shortly after.

Deepgram's core product is Nova-3, its third-generation automatic speech recognition model, which delivers a 47-54% reduction in word error rate compared to competitors and was the first voice AI model to offer self-serve customization, letting developers adapt vocabulary instantly without retraining. Flux, a newer model built specifically for voice agent use cases, adds model-integrated end-of-turn detection and configurable turn-taking dynamics on top of Nova-3-level accuracy, directly addressing one of the hardest problems in conversational AI: knowing when a user has actually finished speaking. Aura-2 handles text-to-speech with 90ms latency for real-time applications, and the Voice Agent API bundles speech-to-text, LLM orchestration, and text-to-speech into a single unified endpoint, removing the need for developers to stitch together three separate services with their own latency and reliability tradeoffs.

Beyond core transcription and synthesis, Deepgram's Audio Intelligence layer adds built-in summarization, sentiment analysis, topic detection, and intent recognition directly on top of transcribed audio, while speaker diarization, punctuation, and custom keyterm prompting round out the platform for production call center and meeting transcription use cases. The company has processed over 50,000 years of audio and transcribed more than 1 trillion words, serving over 1,300 organizations including NASA (transcribing ISS-to-Mission-Control communications), Spotify, Twilio, Citi, and Vapi. In January 2026, Deepgram raised $130 million in a Series C round at a $1.3 billion valuation, becoming the newest unicorn in the voice AI space while remaining cash-flow positive since 2025.

Deepgram's pricing is genuinely usage-based and billed by the exact second with no rounding: new accounts get $200 in free credit with no credit card required, covering tens of thousands of minutes of pre-recorded transcription, before moving to transparent Pay-As-You-Go rates starting around $0.0043/minute for Nova-3 pre-recorded transcription. This makes Deepgram best suited for developers and engineering teams building production voice pipelines, call center analytics, or real-time voice agents who specifically need low latency and high accuracy at scale, rather than non-technical users looking for a simple, no-code transcription dashboard.

Deepgram offers $200 in free credit on signup with no credit card required, usable across any service (Speech-to-Text, Text-to-Speech, Voice Agent API, and Audio Intelligence) and credits do not expire. Pay-As-You-Go pricing for Nova-3 starts at approximately $0.0043/minute for mono pre-recorded transcription (roughly $0.0052/minute for multi-channel/stereo audio) and $0.0077/minute for real-time streaming transcription, with rates dropping to around $0.003/minute at high volume. The Voice Agent API, which bundles speech-to-text, LLM orchestration, and text-to-speech, is priced at $4.50/hour, or roughly $0.08/minute (with a lower bring-your-own-LLM rate around $0.07/minute). Flux text-to-speech is free to use through September 12, 2026 (up to 45 concurrent streaming connections globally, 5 in EU/AU), with standard pricing applying starting September 13, 2026. A Growth tier offers roughly 15-20% discounted rates in exchange for a $4,000+ annual prepayment commitment. On-premise and self-hosted deployment for security, compliance, or latency-sensitive workloads requires a custom Enterprise agreement.