Deepgram API Guide 2026: Nova-3 Speech-to-Text, Pricing, and When to Use It Over Whisper

Whisper is accurate but slow and awkward at scale. Deepgram's Nova models promise faster transcription with lower error rates - and a free credit to test it.

## Deepgram: The Commercial Speech API That Competes With Whisper on Price and Speed Deepgram is a speech-to-text API company whose Nova model family has been a staple for developers since Nova-2 launched in 2023 and Nova-3 followed in 2025. New accounts get $200 in free credits - enough to transcribe roughly 40-60 hours of audio depending on the model tier - which makes it one of the cheapest ways to test a production-grade STT pipeline. ## Where Nova Beats an Open-Source Stack OpenAI's Whisper (106,000+ GitHub stars, MIT license) is the free baseline, and it is excellent. But Deepgram sells conveniences that matter at scale: 1. **Speed.** Nova models transcribe faster than real time, often 3-10x real time, so an hour-long call comes back in minutes. 2. **Streaming with punctuation.** Punctuation, capitalization, and speaker diarization are built in and toggleable, instead of post-processing jobs. 3. **Keyterm prompting.** You can bias recognition toward product names and jargon - crucial for support calls full of API names. 4. **Multilingual coverage.** Nova-3 handles dozens of languages with one API call, and language detection is automatic. ## Pricing Model Deepgram charges per audio minute (or per hour on some plans) and pricing scales with volume; the free tier is a $200 credit rather than a capped number of minutes. For high-volume transcription the per-minute rate drops well below the list price, which is why companies doing tens of thousands of call hours negotiate dedicated plans. ## When to Stay With Whisper If your workload is offline batch transcription, if you need everything on your own hardware for privacy, or if you have GPU capacity sitting idle, Whisper (or whisper.cpp on CPU) is still the right call - free, local, and continuously improved. Deepgram wins when you need low latency, streaming, or scale without managing GPU infrastructure. ## FAQ **Is Deepgram free?** New accounts get $200 in credit; beyond that it is pay-per-minute. **Does Deepgram support text-to-speech too?** Yes, it has TTS models as well, including voice cloning options. **Can I use it for live phone calls?** Yes, it has streaming WebSocket endpoints designed for real-time telephony. **Is my audio stored?** Deepgram offers zero-storage retention options for compliance-sensitive use.