Deepgram is a voice AI platform that exposes speech-to-text, text-to-speech and conversational agents through APIs. Its Flux speech-to-text models listen with built-in turn detection and interruption handling, Flux TTS speaks with latency as low as 80 milliseconds, and a unified Voice Agent API combines transcription, LLM orchestration and speech into one call. Nova-3 models cover 45 or more languages with speaker diarization, smart formatting and keyterm prompting. Usage is billed per minute, with a $200 starting credit and a Growth tier from $4K per year.
Developers and product teams building voice agents, call center software and transcription features are the core audience. Startups can sign up without a credit card and pay as they go, growing applications can pre-pay credits for savings of up to 20%, and enterprises with compliance or custom model needs can talk to sales. Platforms and partners can embed the technology in their own products, and teams that need data control can self-host.
Typical projects include real-time voice agents that handle interruptions naturally, live captioning and meeting transcription, and analysis of recorded conversations through audio intelligence. Streaming and pre-recorded audio are priced separately, with Nova-3 monolingual streaming currently at $0.0048 per minute against a regular $0.0077. Concurrency limits scale by plan, with up to 50 simultaneous REST speech-to-text requests and up to 150 on the WebSocket API for pay-as-you-go accounts. Support runs through community and Discord channels unless an enterprise agreement is in place. Developers can try the playground in the browser before writing code, and the Growth tier suits teams whose usage has become predictable enough to pre-pay for the year.
