whatmodel.sh / stt / Deepgram Nova 2
Deepgram Nova 2
↗ share
+ contribute
Deepgram Nova 2 is a closed-weights speech-to-text model from Deepgram, released September 2023.
Catalogue identity notes
Deepgram API model=nova-2. Keep distinct from Nova-3.
Model information sources
Score within the Open ASR English short-form WER (March 2026) comparison set only.
$ how it compares
The speech-to-text models ranked either side of Deepgram Nova 2, out of 176 with a score, then more from Deepgram.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 41 |  Granite Speech 3.3 2B | 60±3 | local | ||
| 42 |  Speechmatics Enhanced | 60±2 | api $0.0067/audio min | ||
| 43 | Deepgram Nova 2 | 60±4 | api $0.0058/audio min | ||
| 44 |  Fun-ASR-Flash | 59±4 | comparison only | ||
| 45 | Granite Speech 3.3 8B | 59±5 | local | ||
| more from Deepgram | |||||
| 27 | Nova-3 | 61±2 | api $0.0052/audio min | ||
see all speech-to-text models · find the ones that fit your machine
$ what it's good at
other tests5 not in a task · 93% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Russian STT · podlodka corpus WER (20-clip pilot) | 9.3% ✎ | █··········▄ beats 1/2 | 34% |
| Russian STT · rulibrispeech corpus WER | 10.2% ✎ | █··█·······█ beats 1/2 | 34% |
| Russian STT · fleurs corpus WER | 4.0% ✎ | █··█·······█ beats 2/2 | 26% |
| Voice memos · RU 10-clip pilot median WER | 21.4% ✎ | █··········█ beats 1/1 | — |
| Voice memos · EN 10-clip pilot median WER | 20.5% ✎ | █··········█ beats 1/1 | — |
$ where to run it
Through an API
api · $0.0058/audio min · deepgram.com ↗ · price sourceUSD per input audio minute: Deepgram states Nova-2 remains available at unchanged rates, 'Nova-2 streaming at $0.35/hour' (÷60). This is the only Nova-2 rate on the page (streaming); a separate pre-recorded Nova-2 rate is not published there. Product: Deepgram STT API. API model: nova-2. Settings: pay-as-you-go, streaming. Price basis: published list rate, per-hour to per-minute scaling only. Exclusions: Growth-plan discount (~12.5%), add-ons. Currency: USD. Deepgram official pricing page, observed 2026-09-25.
$ how the score was worked out · Russian STT benchmark by doma… 93%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Russian STT benchmark by domain | 3 | 93% |
| prior · pull toward the middle | 7% | |
| weighted average | = 59.7 |
not counted (2)
| Live voice-memo STT pilot | Voice memos · RU 10-clip pilot median WER [correct] | 21.4% | fewer than three models have this test yet, or this source ran it for this model alone |
| Live voice-memo STT pilot | Voice memos · EN 10-clip pilot median WER [correct] | 20.5% | fewer than three models have this test yet, or this source ran it for this model alone |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json