Deepgram Nova-3
Deepgram
Deepgram's speech-to-text model (January 2025), the replacement for Nova-2, delivering large word-error-rate reductions (54.3% streaming, 47.4% batch) versus competitors. Nova-3 is the first voice AI model with real-time multilingual transcription and self-serve keyterm customization without retraining.
Strengths
- Industry-leading accuracy (54.3% streaming WER reduction vs peers)
- First real-time multilingual transcription with code-switching
- Self-serve keyterm prompting without model retraining
- Very low cost (~$0.0043/min pre-recorded)
- Streaming and batch, with enterprise/on-prem options
Caveats
- Proprietary - API access (on-prem for enterprise)
- Transcription only (paired with separate TTS for voice agents)
- Accuracy varies by language and audio quality
- Vendor
- Deepgram
- License
- Proprietary
- Release Date
- 2025-01-01
- Modalities
- audiotext
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
~$0.0043/min pre-recorded (~$0.26/hr) and ~$0.0077/min streaming via the Deepgram API; enterprise and on-prem options available.