Deepgram Nova-2
Deepgram Nova-2 is a previous general ASR SKU, not the 2026 default. Rechecked 2026-08-17: Deepgram models overview lists Flux for voice-agent pipelines and Nova-3 as our highest-performing general-purpose ASR. Nova-2 remains listed for languages Nova-3 does not yet cover and for filler-word identification. Drop leftover fastest commercial STT / <300ms / 40x realtime / WER 5-8% as an audited 2026 crown.
Compare Whisper V3 if you wanted OpenAI's open-weight family, and the live Deepgram docs if you wanted Flux or Nova-3 instead of this snapshot.
Key Features
- Still in the catalog, not the lead: Docs keep nova-2 / nova-2-general as an option. Recheck the language table before you pick it.
- Nova-3 and Flux are the current recommendations: Flux is described as ultra-low latency for voice agents with Nova-3-level accuracy. Nova-3 is the general ASR lead.
- No invented latency or WER: Average 250ms and 5-8% WER are out unless you re-run a first-party card.
- No invented dollar: We did not open a 2026 per-minute Deepgram table in this pass.
Use Cases
- People whose language or filler-word need still maps to Nova-2 on the docs table.
- People comparing leftover Nova-2 to Nova-3 / Flux.
- People who were about to paste 40x realtime into a 2026 deck.
Limitation: We did not invent a Nova-3 page on this site. Recheck Deepgram docs for the live SKU.
Pricing
Deepgram docs on 2026-08-17. No public dollar matrix on the overview we opened.
| Piece | Price | Notes from that page |
|---|---|---|
| Nova-2 | Previous general ASR | Still listed. Use when Nova-3 lacks the language or you need filler words. |
| Nova-3 / Flux | Recheck Deepgram pricing | Current recommended general ASR / voice-agent paths. |
If a leftover blog still says Nova-2 is the fastest 2026 model, treat this page as the correction.
Getting Started
- Open the models overview.
- Prefer Flux or Nova-3 unless the table sends you to Nova-2.
- Recheck Whisper V3 if you wanted open weights.
- Do not paste 40x / WER 5-8% into a deck.
Frequently Asked Questions
Is Nova-2 still the default?
No. Docs recommend Nova-3 for general ASR and Flux for voice agents.
Why is Nova-2 still listed?
Languages not yet on Nova-3, and filler-word identification.
How much is it?
Not on the overview we opened. Recheck Deepgram pricing.
Alternatives
- Whisper V3: MIT open-weight family.
- whisper-1: Older OpenAI API object.
- OpenAI: Live hosted catalog if you wanted API STT.
Tips
- Call Nova-2 a previous SKU.
- Quote Nova-3 / Flux as the docs lead.
- Recheck the language table before you force Nova-2.
Conclusion
Deepgram Nova-2 is a previous ASR option still listed for some languages, not the 2026 fastest crown. Start at the models overview, then decide whether Nova-3 or Flux already covers the job.
Comments
No comments yet. Be the first to comment!
Related Tools
Whisper V3
github.com/openai/whisper
Open-weight Whisper large-v3 family. Rechecked: MIT 107,437 stars. Not OpenAI's current hosted STT default.
Mistral Voxtral Mini Transcribe Realtime
docs.mistral.ai/models
Mistral Voxtral Mini Transcribe Realtime: live realtime transcription sibling. Rechecked 2026-08-18. Not leftover Transcribe 25.07. No invented dollar row.
Claude 3.5 Sonnet
docs.claude.com/en/docs/about-claude/models/overview
Previous Claude 3.5 Sonnet snapshot. Rechecked: live docs lead with Claude 5 / 4.6. Not the 2026 default.
Related Insights

Grok Bot and Hermes Bot: one person finally gets a think tank and a secretariat
Grok Bot now ships with Cursor Pro+. Hermes Bot runs on a VPS. They are not smarter chat boxes. The think tank advises, the secretariat executes, and you still make the call.
Hook OpenCode Go into Codex on Windows. Do Not Open a Second Toolkit.
A ChatGPT-signed Codex desktop app still shows mostly GPT in the picker. On Windows, enable only OpenCode Go and the Grok, GLM, Kimi, DeepSeek, and MiniMax models you already pay for appear in the same selector. Keys stay local. Native GPT stays put.
What Locks Codex Is the Picker, Not the Models
You already pay for OpenCode Go, Grok, and Z.ai, but the Codex picker still shows mostly GPT. The community project codex-router does not teach another install ritual. It puts subscriptions you already bought back into the selector, keeps keys on the machine, and leaves native GPT alone.