Deepgram Nova-3
Deepgram Nova-3 is 4th of 17 on final-text delay and 12th of 17 on the Pipecat Dataset.
Last updated Methodology by Dileep Chagam
- 3.46%
- Pipecat WER · 12th
- 4.39%
- Ocular WER · 11th
- 1.06s
- First text · 7th
- 103ms
- Final text · 4th
Likely range, with 95% confidence: Pipecat WER 3.16 to 3.78%, Ocular WER 3.42 to 5.25%.
Accuracy vs speed
Each dot is one model. Bottom-left is best: fewer errors and a faster final transcript.
Deepgram Nova-3 against every other model
Open on the metric Deepgram Nova-3 places highest on, with the other three a tab away. Lower is better on all of them, so the order runs best first.
Final-text delay after the speaker stops. Lower is better.
| # | Model | Final text | turns | |
|---|---|---|---|---|
| 1 | Soniox STT-RT v5 | 38ms | 200 | |
| 2 | Inworld STT-1 | 41ms | 206 | |
| 3 | AssemblyAI Universal 3.6 Pro | 91ms | 206 | |
| 4 | Deepgram Nova-3 | 103ms | 202 | |
| 5 | Deepgram Flux English | 108ms | 199 | |
| 6 | Deepgram Flux Multilingual | 114ms | 175 | |
| 7 | Cartesia Ink 2 | 123ms | 198 | |
| 8 | Speechmatics Linden | 147ms | 194 | |
| 9 | AssemblyAI Universal 3.5 Pro | 180ms | 205 | |
| 10 | Smallest Pulse | 201ms | 206 | |
| 11 | Gradium | 212ms | 198 | |
| 12 | Reson8 | 235ms | 201 | |
| 13 | Google Chirp 3 | 475ms | 206 | |
| 14 | GPT Realtime Whisper | 543ms | 206 | |
| 15 | Google Chirp 2 | 568ms | 205 | |
| 16 | GPT-4o Mini Transcribe | 630ms | 206 | |
| 17 | GPT-4o Transcribe | 652ms | 206 |
Median across the same 206 Ocular Dataset turns. Providers finalize differently, so compare models with similar finalization contracts before treating small gaps as meaningful.
- WER · Word error rate
- Share of words the model got wrong: replaced, missing or added. Lower is better.
- Pipecat Dataset
- 1,000 short public clips from Pipecat's STT benchmark dataset.
- Ocular Dataset
- 8 recordings from 4 real phone conversations, licensed from Ocular.
- Likely range · 95% range
- Where the model's true score probably falls, given how much audio was tested. Overlapping ranges are too close to call.
- TTFT · Time to first text
- How long after the audio starts streaming until the first words appear.
- TTFS · Final-text delay
- How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.
First text and final text, at three percentiles
Measured on the same 206 streamed Ocular Dataset turns as every other model. P90 and P95 say what a caller meets on a bad turn, which a median hides.
| Measure | P50 | P90 | P95 |
|---|---|---|---|
| Time to first text | 1.06s | 1.10s | 2.07s |
| Final-text delay | 103ms | 143ms | 155ms |
- P50 / P90 / P95 · Typical, slow and slowest turns
- Half of turns are faster than P50. Only 1 in 10 is slower than P90, and 1 in 20 slower than P95.
Deepgram Nova-3 on the other two views
The same charts the benchmark page carries, with Deepgram Nova-3 ringed. Consistency across the two datasets is one reading; how fast a model starts against how fast it commits is the other.
Public clips vs real calls
Each dot is one model. Bottom-left is best on both datasets.
First words vs final transcript
Each dot is one model. Bottom-left is fastest at both.
How much audio it was tested on
- 1,000
- Pipecat clips scored
- 8
- Ocular recordings scored
- 202
- turns timed
- 0
- requests that failed
A clip counts only where the model returned usable text, so counts differ between models. Word error rate is measured the same way throughout: incorrect, missing, and extra words against the reference transcript.
Deepgram Nova-3 head to head
- vs AssemblyAI Universal 3.6 Pro
- vs AssemblyAI Universal 3.5 Pro
- vs Google Chirp 3
- vs Reson8
- vs GPT Realtime Whisper
- vs Cartesia Ink 2
- vs Speechmatics Linden
- vs Soniox STT-RT v5
- vs Google Chirp 2
- vs Inworld STT-1
- vs GPT-4o Mini Transcribe
- vs Smallest Pulse
- vs GPT-4o Transcribe
- vs Deepgram Flux English
- vs Deepgram Flux Multilingual
- vs Gradium
Frequently asked questions
Where does Deepgram Nova-3 place in the benchmark?
Deepgram Nova-3 places highest on final-text delay, 4th of 17 at 103ms. Its weakest placing is the Pipecat Dataset, 12th of 17 at 3.46%.
How accurate is Deepgram Nova-3?
Deepgram Nova-3 transcribes the 1,000 public Pipecat clips at 3.46% word error rate and the eight licensed Ocular recordings at 4.39%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.
How fast is Deepgram Nova-3 in a live conversation?
First text arrives after 1.06s at the median, and the transcript is final 103ms after the speaker stops. Both are measured on the same 206 streamed turns as every other model, and the page shows P90 and P95 as well, which is what a caller meets on a bad turn.
How was Deepgram Nova-3 measured?
Every model runs the same two datasets under the same conditions: 1,000 public clips from Pipecat's STT benchmark dataset for accuracy at scale, and eight licensed conversation recordings from Ocular for real speech. Latency comes from 206 streamed turns, using each provider's own supported finalization contract. Deepgram Nova-3 returned usable text for 1,000 of the public clips.