Skip to content

AI model by Deepgram

Deepgram Nova-3 and Flux for business

Deepgram's speech-to-text models: Nova-3 for general transcription in 60+ languages, and Flux for voice agents with built-in end-of-turn detection.

  • Reviewed on September 24, 2026
  • Deepgram

Capability tiersRelative, not benchmarks

Reasoning Very low
Coding Very low
Vision Very low
Speed Very high
Price level Low
Context size Very low
Tool calling Very low
Structured output Very low

Key facts about Deepgram Nova-3 and Flux

Model ID at review
nova-3, flux-general-en, flux-general-multi
Provider
Deepgram
Main category
Speech
Open weights
No, available as a hosted service
Last reviewed
September 24, 2026

Fit

Where it fits and where it does not

Good at

  • Real-time streaming transcription.

  • Voice agents, with end-of-turn detection in Flux.

  • Many languages, including switching mid-conversation in Nova-3.

  • Custom vocabulary without retraining.

  • Specialized medical versions of Nova-3.

Not the right choice for

  • Generating speech, which needs a text-to-speech model.

  • Summaries or answers on their own.

  • Recording people without consent.

  • Uses that need a transcript with no human review.

Use cases

Business use cases we would use it for

Phone assistants

The listening part of a voice assistant that knows when the caller has finished. See voice AI.

Call and meeting notes

Transcripts that feed summaries and follow-ups. See meeting notes.

Call center analysis

Transcribing support calls for search, quality review and training.

Our notes

When we would choose it

Deepgram offers two main speech-to-text lines at the time of review. Nova-3 is its general-purpose model, covering more than 60 languages, handling speakers who switch languages mid-conversation and accepting custom vocabulary without retraining. There are specialized medical and pharmaceutical versions. Flux is newer and built for voice agents, with end-of-turn detection inside the model. Flux Multilingual became generally available in April 2026 with 10 languages.

End-of-turn detection matters more than it sounds. A voice assistant that answers too early interrupts people, and one that waits too long feels slow. Building turn detection into the speech model makes that timing more natural.

Both models work for streaming and for pre-recorded audio, through REST and WebSocket APIs. Deepgram also launched its own text-to-speech in August 2026, which we have not reviewed here.

For transcription of recorded meetings, compare Nova-3 with OpenAI GPT Transcribe. For live voice agents, Flux is a strong candidate. Pair either with a language model and a voice such as ElevenLabs or OpenAI TTS.

Before launching a voice assistant, we would run test calls with real staff and customers, measure how often the assistant interrupts or waits too long, and check that callers are told they are speaking with an AI. See AI customer support.

Before you commit

Things to check before you commit

  • Your own audio

    Test real calls with your accents, noise and terms.

  • Languages

    Nova-3 covers 60+ languages. Flux Multilingual covers 10. Check yours.

  • Consent and data

    Get consent to record and check data terms for audio.

  • Deployment

    Deepgram mentions self-hosted releases. Ask if you need on-premise.

Alternatives

Models to compare it with

GPT Transcribe

OpenAI's current speech-to-text model, replacing Whisper and the GPT-4o transcribe models, with streaming and keyword or language hints.

ElevenLabs Eleven v3

ElevenLabs text-to-speech: the expressive Eleven v3 in 70+ languages, a real-time conversational version, and the low-latency Flash models, with voice cloning.

GPT-4o mini TTS

OpenAI's current text-to-speech model for turning text into natural spoken audio, at a low price level.

Keep exploring

FAQ

Questions people ask us

Have a question that is not here? Ask us directly.

Start a project

Not sure which model fits? Ask us to evaluate your use case.

Send a short brief. We reply with questions, a suggested plan and an estimate you can compare with other offers.

Your privacy choices

We use necessary storage to run this site. With your permission we also use Google Analytics to see which pages help people, and load maps from Google. You can change this at any time. Read the cookie policy.