Deepgram Pricing in May 2026

Deepgram's pricing follows usage-based per-minute model: $0.0043/minute for Nova-3 (general), $0.0145/minute for Nova-3 Medical (specialized), real-time transcription at similar rates. Real-world Deepgram bills for AI products with transcription land at $100-1,000/month for early-stage; mid-stage products with high-volume voice workloads hit $5K-50K/month at scale.

The Free $200 Credits offer is genuinely substantial — covers approximately 46K minutes of transcription. The strategic decision: Deepgram's value is the latency-and-accuracy combination: their custom acoustic models are tuned for ultra-low-latency real-time use cases (voice agents, live captioning) while maintaining accuracy. For broader speech-to-text APIs, AssemblyAI competes directly with broader audio-intelligence features (sentiment, topics, summarization). OpenAI Whisper API targets transcription via OpenAI at competitive pricing — usually preferred for batch use cases due to higher latency. Google Cloud Speech-to-Text targets enterprise. Azure Speech Services and AWS Transcribe target their respective clouds. Rev.ai targets enterprise transcription. For specifically open-source self-hosted, Whisper (OpenAI's open-weight model) and faster-whisper run locally or on GPUs. WhisperX adds diarization. For specifically real-time voice agents with low-latency, Deepgram is uniquely positioned for sub-300ms transcription latency that voice agents require. Vapi, Retell AI, and Bland AI build voice-agent infrastructure on top of Deepgram. For specifically text-to-speech (Deepgram's recent expansion), ElevenLabs is the dominant alternative. For specifically meeting transcription (different use case), Otter.ai, Fireflies.ai, and tldv target meeting-recording-and-transcription. Deepgram's specific niche is ultra-low-latency real-time STT for voice agents and live captioning.