Deepgram (deepgram.com) is the speech AI infrastructure layer powering the majority of production voice agent stacks in 2026. Its Nova-3 speech-to-text model delivers best-in-class accuracy with 90ms Aura-2 TTS latency, making it the default ASR choice for real-time pipelines built on platforms like Vapi and Retell AI. Beyond transcription, Deepgram has expanded into a Voice Agent API that bundles STT, LLM, and TTS into a single $0.08/min endpoint — removing the need to wire three separate services together.
Key Features
- Nova-3 STT — state-of-the-art accuracy with speaker diarization, punctuation, and custom vocabulary
- Aura-2 TTS — 90ms latency text-to-speech optimized for real-time voice agent responses
- Voice Agent API — bundled STT + LLM + TTS at $0.08/min; bring-your-own LLM option at $0.07/min
- Audio Intelligence — built-in summarization, topic detection, sentiment analysis, and intent recognition
- Streaming & batch — real-time WebSocket and REST pre-recorded transcription on the same API
- On-premise deployment — enterprise option for security, compliance, or latency requirements
Deepgram Pricing

- Free — $200 in credits — Full API access, no credit card required. Credits cover all endpoints.
- Pay-As-You-Go — STT from $0.0043/min pre-recorded, $0.0077/min streaming. TTS at $30/1M characters. Voice Agent API at $4.50/hour.
- Growth — $4,000+/year prepaid — Up to 20% lower rates vs. pay-as-you-go. Higher concurrency limits.
- Enterprise — custom pricing — Custom rates, on-premise deployment, dedicated SLAs, compliance support.Pricing is subject to change. Always check the latest rates on the official website. For more AI tool reviews, visit aitoolscoop.com.
Who Should Use Deepgram?
Deepgram is the right STT layer for any team building production voice agents, call center automation, meeting transcription, or medical documentation tools. Its Nova-3 accuracy and Aura-2 latency lead the ASR market, and the Voice Agent API is the fastest way to get a complete voice pipeline running with a single vendor. Teams building on Vapi or Retell AI typically use Deepgram as their default STT provider.