Voice AI BenchmarksBy

DEEPGRAMSPEECH-TO-TEXT130,016 SAMPLES / 30 DAYSLAST RUN OCT 14, 2026, 02:30 UTC

official resource

Nova 3 speech-to-text benchmarks

Nova 3, hosted by Deepgram, measures mean 85 ms time to final segment (5th of 29) and 5.3% word error rate (20th of 31) among STT systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

Nova 3 is Deepgram's current general-purpose speech-to-text model.

Time to Final Segment#5 / 29
85ms
Word Error Rate#20 / 31
5.3%
Time to First Token#11 / 27
1431ms

Overview

Nova 3 handles multilingual audio in real time, and the tested endpoint adds diarization, code switching and keyterm controls.

Nova 3 is tested every day on fixed public audio — clean, accented, noisy, reverberant, far-field, clipped and phone-codec speech — for transcription accuracy and streaming latency.

Technical specifications

Made by
Deepgram
Hosted by
Deepgram
Source
Official API
Licensing
Proprietary
Deployment
On-prem
Region
US
Features
Code switching, Diarization, Keyterm biasing, Multilingual, VAD

How Nova 3 ranks

Full STT dashboard
Time to Final Segment — every measured STT modelMilliseconds · lower is better · 30-day averageEvery STT model ranked on Time to Final Segment, with Nova 3 highlighted.
  1. #1Qwen3 ASR 1.7b39 ms
  2. #2STT RT v545 ms
  3. #3Qwen3 ASR Fast47 ms
  4. #4STT 166 ms
  5. #5Nova 385 ms
  6. #6Nova 287 ms
  7. #7Parakeet TDT 0.6B v397 ms
  8. #8Flux Multilingual98 ms
  9. #9Flux99 ms
  10. #10Scribe v2 Realtime122 ms
  11. #11Whisper Large v3via Baseten131 ms
  12. #12Universal 3.6 Pro136 ms
Show all 29 models
  1. #13Ink 2148 ms
  2. #14Universal 3.5 Pro167 ms
  3. #15Whisper Large v3via Together AI200 ms
  4. #16Pulse205 ms
  5. #17Grok STT224 ms
  6. #18resonant-1252 ms
  7. #19Default257 ms
  8. #20Linden 1270 ms
  9. #21Gemini 3.5 Transcribe Live317 ms
  10. #22Daytona STT399 ms
  11. #23Voxtral Mini Transcribe Realtime 2602410 ms
  12. #24GPT Realtime Whisper599 ms
  13. #25Solaria 1640 ms
  14. #26GPT-4o mini Transcribe707 ms
  15. #27Chirp 3768 ms
  16. #28GPT-4o Transcribe783 ms
  17. #29Chirp 2908 ms
median of all models · 200 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Qwen3 ASR 1.7bBaseten39 ms945 ms13,270
2STT RT v5Soniox45 ms1518 ms23,533
3Qwen3 ASR FastNari Labs47 ms1752 ms26,652
4STT 1Inworld AI66 ms1214 ms24,749
5Nova 3Deepgram85 ms1431 ms26,002
6Nova 2Deepgram87 ms1417 ms25,960
7Parakeet TDT 0.6B v3Together AI97 ms1229 ms26,460
8Flux MultilingualDeepgram98 ms1165 ms26,007
9FluxDeepgram99 ms1091 ms26,005
10Scribe v2 RealtimeElevenLabs122 ms2158 ms26,640
11Whisper Large v3Baseten131 ms923 ms13,264
12Universal 3.6 ProAssemblyAI136 ms1072 ms24,052
13Ink 2Cartesia148 ms1965 ms26,652
14Universal 3.5 ProAssemblyAI167 ms1025 ms26,654
15Whisper Large v3Together AI200 ms1262 ms26,445
16PulseSmallest205 ms1976 ms26,654
17Grok STTSpaceXAI224 ms—26,576
18resonant-1Reson8252 ms—26,653
19DefaultGradium257 ms1986 ms26,632
20Linden 1Speechmatics270 ms1537 ms26,595
21Gemini 3.5 Transcribe LiveGemini317 ms1610 ms26,567
22Daytona STTGuava399 ms5700 ms4,794
23Voxtral Mini Transcribe Realtime 2602Mistral410 ms1852 ms26,199
24GPT Realtime WhisperOpenAI599 ms1824 ms26,622
25Solaria 1Gladia640 ms1786 ms26,195
26GPT-4o mini TranscribeOpenAI707 ms—26,621
27Chirp 3Google768 ms5957 ms26,638
28GPT-4o TranscribeOpenAI783 ms—26,625
29Chirp 2Google908 ms6098 ms26,650
—Nemotron 3.5 ASR StreamingTogether AI—1544 ms26,542
—Universal StreamingAssemblyAI—1508 ms26,623
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Latency vs accuracy

TTFS vs WEREach point is one measured model · Nova 3 highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Nova 3 highlighted.
Qwen3 ASR 1.7b — 39 ms, 2.9%STT RT v5 — 45 ms, 4.8%Qwen3 ASR Fast — 47 ms, 2.6%STT 1 — 66 ms, 7.0%Nova 2 — 87 ms, 6.7%Parakeet TDT 0.6B v3 — 97 ms, 11.1%Flux Multilingual — 98 ms, 7.5%Flux — 99 ms, 5.8%Scribe v2 Realtime — 122 ms, 4.4%Whisper Large v3 via Baseten — 131 ms, 4.5%Universal 3.6 Pro — 136 ms, 2.1%Ink 2 — 148 ms, 3.6%Universal 3.5 Pro — 167 ms, 2.4%Whisper Large v3 via Together AI — 200 ms, 7.4%Pulse — 205 ms, 4.6%Grok STT — 224 ms, 4.1%resonant-1 — 252 ms, 2.6%Default — 257 ms, 8.8%Linden 1 — 270 ms, 3.9%Gemini 3.5 Transcribe Live — 317 ms, 3.0%Daytona STT — 399 ms, 5.0%Voxtral Mini Transcribe Realtime 2602 — 410 ms, 5.4%GPT Realtime Whisper — 599 ms, 4.0%Solaria 1 — 640 ms, 6.1%GPT-4o mini Transcribe — 707 ms, 3.9%Chirp 3 — 768 ms, 3.2%GPT-4o Transcribe — 783 ms, 3.8%Chirp 2 — 908 ms, 3.9%Nova 3 — 85 ms, 5.3%

Where the errors come from

WER compositionNova 3's Word Error Rate split by error type · 30-day averageNova 3's WER split into substitutions, deletions and insertions.
  • Nova 35.3%
SubstitutionsDeletionsInsertions

Averages and tail latency

Averages hide slow outliers — these are the distributions behind each figure.

Nova 3 latency distributionMilliseconds · shared axis across metrics · last 30 daysNova 3's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to Final Segmentp50 90 ms · Max 19227 ms
  • Time to First Tokenp50 997 ms · Max 5435 ms
band p25–p75 · tick p50 · whisker to max with p90 and p95 stops
Average and percentile values per metric, with the number of samples behind each row.
MetricAveragep25p50p75p90p95MaxSamples
Time to Final Segment85 ms55 ms90 ms103 ms121 ms134 ms19227 ms26,002
Word Error Rate5.3%0.0%3.1%8.3%15.3%20.0%100.0%26,005
Time to First Token1431 ms979 ms997 ms1979 ms1989 ms2005 ms5435 ms26,005

Last 30 days

Daily medians from the same measurement runs · gaps are days without qualifying runs.

Time to Final Segment — daily p50Line p50 · band p25–p75 · UTC daysNova 3's daily median Time to Final Segment over the last 30 days.
Nova 3 · Sep 14: 91 msNova 3 · Sep 15: 94 msNova 3 · Sep 16: 78 msNova 3 · Sep 17: 87 msNova 3 · Sep 18: 83 msNova 3 · Sep 19: 76 msNova 3 · Sep 20: 73 msNova 3 · Sep 21: 91 msNova 3 · Sep 22: 85 msNova 3 · Sep 23: 87 msNova 3 · Sep 24: 95 msNova 3 · Sep 25: 93 msNova 3 · Sep 26: 61 msNova 3 · Sep 27: 94 msNova 3 · Sep 28: 62 msNova 3 · Sep 29: 86 msNova 3 · Sep 30: 93 msNova 3 · Oct 1: 84 msNova 3 · Oct 2: 92 msNova 3 · Oct 3: 72 msNova 3 · Oct 4: 81 msNova 3 · Oct 5: 89 msNova 3 · Oct 6: 86 msNova 3 · Oct 7: 95 msNova 3 · Oct 8: 88 msNova 3 · Oct 9: 91 msNova 3 · Oct 10: 68 msNova 3 · Oct 11: 83 msNova 3 · Oct 12: 80 msNova 3 · Oct 13: 95 msNova 3 · Oct 14: 86 ms
Word Error Rate — daily averageDaily average · UTC daysNova 3's daily Word Error Rate over the last 30 days.
Nova 3 · Sep 14: 6.3%Nova 3 · Sep 15: 6.0%Nova 3 · Sep 16: 7.4%Nova 3 · Sep 17: 7.0%Nova 3 · Sep 18: 6.4%Nova 3 · Sep 19: 5.1%Nova 3 · Sep 20: 5.6%Nova 3 · Sep 21: 5.9%Nova 3 · Sep 22: 5.7%Nova 3 · Sep 23: 5.6%Nova 3 · Sep 24: 6.0%Nova 3 · Sep 25: 7.1%Nova 3 · Sep 26: 6.1%Nova 3 · Sep 27: 5.6%Nova 3 · Sep 28: 5.0%Nova 3 · Sep 29: 5.6%Nova 3 · Sep 30: 5.8%Nova 3 · Oct 1: 6.8%Nova 3 · Oct 2: 4.8%Nova 3 · Oct 3: 6.3%Nova 3 · Oct 4: 7.8%Nova 3 · Oct 5: 6.7%Nova 3 · Oct 6: 6.2%Nova 3 · Oct 7: 4.9%Nova 3 · Oct 8: 6.6%Nova 3 · Oct 9: 5.2%Nova 3 · Oct 10: 5.5%Nova 3 · Oct 11: 6.4%Nova 3 · Oct 12: 4.8%Nova 3 · Oct 13: 6.2%Nova 3 · Oct 14: 13.5%

Time to Final Segment by dataset

Nova 3 by test conditionMilliseconds · lower is better · best condition firstNova 3's Time to Final Segment on each benchmark dataset.
  1. WildASR accents82 ms
  2. PipeCat (production)83 ms
  3. WildASR noise gaps86 ms
  4. WildASR far-field88 ms
  5. WildASR clean88 ms
  6. WildASR clipping91 ms
  7. WildASR phone codec91 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
ProductionPipeCat83 ms13,678
AccentsWildASR82 ms1,369
CleanWildASR88 ms5,475
ClippingWildASR91 ms1,371
Far-fieldWildASR88 ms1,369
Noise gapsWildASR86 ms1,369
Phone codecWildASR91 ms1,371

Strongest condition: WildASR accents at 82 ms · weakest: WildASR phone codec at 91 ms.

How fast is Nova 3?

On Deepgram, Nova 3 measures mean 85 ms time to final segment (5th of 29) and mean 1431 ms time to first token (11th of 27). Last measured 2026-10-14.

How accurate is Nova 3?

On Deepgram, Nova 3 measures 5.3% word error rate (20th of 31). Last measured 2026-10-14.

Who hosts Nova 3?

Nova 3 is created by Deepgram and served by Deepgram. Coval measures each hosted endpoint separately.

Limits of this comparison

Coval reports separate results for clean, accented, noisy, reverberant, far-field, clipped and phone-codec audio.

  • Language, endpointing and keyword settings can change production outcomes beyond Coval's reproducible configuration.

Official sources

Results are re-measured daily using fixed datasets and reported over a rolling 30-day window. Same datasets, prompts and metric definitions for every model, measured by Coval’s open-source runner. Full methodology on the overview.

Evaluate your own voice agent

Use Coval to test your production configuration, prompts and calls—not only the public benchmark endpoints.

Book a Demo