LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Deepgram Nova 3
by Deepgram
CLOSEDHOSTEDAPI
RANKED #249 OF 326 ASSISTANTS · OVERALL #390 OF 7,057 · VIBE SCORE 5.4 · 30 VOICES
General-purpose chat model for instruction following, writing, and analysis
HAPPY8.8
4 mentions
SAD6.3
1 mentions
VIBE TRAJECTORY
26 weeks · 36 voices
HONEYMOON FADING
Early praise is cooling off in recent weeks.
PREFER OPEN SOURCE? THE CROWD HAS PICKS
Deepgram Nova 3 is closed, so you rent it. These open-weight models can genuinely stand in for it. Download once, own forever.
WHAT THE INTERNET SAYS
People are mostly frustrated with Deepgram Nova 3 right now.
26 POSTS · 4 COMMENTS · GITHUB · X · REDDIT
PEOPLE LOVE
- No fan mail yet. The crowd is still warming up.
PEOPLE COMPLAIN ABOUT
- Sometimes makes things up (3 mentions)
TOPIC BY TOPIC
- Getting facts rightA SORE SPOT
0 thumbs up · 3 thumbs down
LOOKING FOR AN ALTERNATIVE?
Other models the crowd has fully reviewed, starting with text models like this one.
Muse Spark 1.3> SOURCES (12)
- REDDIT · 2025-03-04 · 2 UPVOTESDeepgram Nova-3 Medical
Deepgram Nova-3 Medical
- GITHUB · 2026-08-26 · 1 UPVOTESfeat: managed live voice runs Flux, with an opt-out on Settings > Voice
feat: managed live voice runs Flux, with an opt-out on Settings > Voice
- GITHUB · 2026-09-23 · 1 UPVOTESfeat(asr): switch summarization and enhancement to Gemma 4 26B, usage-based billing, and tighter Whisper hallucination guards
Summary Replaces Llama 3.2 3B with Google Gemma 4 26B A4B (@cf/google/gemma-4-26b-a4b-it) for summaries and text enhancement, and tightens Whisper's non-English transcription to reduce hallucinated text. Changes - Switches the text generation model from…
- X · 2026-08-18 · 1 UPVOTESDay 13 AI Engineer. Today I shipped more than I consumed: • Finished last week's n8n AI agent — RAG email assistant +
Day 13 AI Engineer. Today I shipped more than I consumed: • Finished last week's n8n AI agent — RAG email assistant +
- GITHUB · 2026-09-22 · 1 UPVOTESperf(auto-answer): cut time-to-first-answer — prefetch that paints, measured judge, 'none' reasoning
perf(auto-answer): cut time-to-first-answer — prefetch that paints, measured judge, 'none' reasoning
- GITHUB · 2026-08-13 · 0 UPVOTESfix(apps): make the cloud Provider and Model dropdowns consistent
fix(apps): make the cloud Provider and Model dropdowns consistent
- GITHUB · 2026-08-25 · 0 UPVOTESfeat(assistant): reach managed Flux through a model family, not a new provider id (JARVIS-1637)
feat(assistant): reach managed Flux through a model family, not a new provider id (JARVIS-1637)
- GITHUB · 2026-08-13 · 0 UPVOTEStranscribe: nova-3, vocab boosting, filler removal, Claude polish
transcribe: nova-3, vocab boosting, filler removal, Claude polish
- X · 2026-08-15 · 0 UPVOTES12/ the stack, for the builders: 🎙️ Murf Falcon: TTS, ultra-low latency 👂 Deepgram Nova 3: multilingual STT 🧠 Gemini
12/ the stack, for the builders: 🎙️ Murf Falcon: TTS, ultra-low latency 👂 Deepgram Nova 3: multilingual STT 🧠 Gemini
- GITHUB · 2026-08-31 · 0 UPVOTESRELEASE: Video transcription on Deepgram Nova-3 (drops Groq Whisper); LLM metadata falls back to Grok
RELEASE: Video transcription on Deepgram Nova-3 (drops Groq Whisper); LLM metadata falls back to Grok
- GITHUB · 2026-08-26 · 0 UPVOTESSupport Deepgram Flux (streaming-only /v2/listen model), and keep streaming-only models out of the transcription picker
Support Deepgram Flux (streaming-only /v2/listen model), and keep streaming-only models out of the transcription picker
- GITHUB · 2026-08-26 · 0 UPVOTESvoice: Flux migration, server-owned turn identity, and the fixes
voice: Flux migration, server-owned turn identity, and the fixes