Voice AI BenchmarksBy

SMALLESTSPEECH-TO-TEXT146,729 SAMPLES / 30 DAYSLAST RUN AUG 28, 2026, 21:30 UTC

official resource

Pulse speech-to-text benchmarks

Pulse is Smallest.ai's real-time speech recognition model.

Time to Final Segment#13 / 24
205ms
Word Error Rate#12 / 28
5.3%
Time to First Token#21 / 24
2072ms

Overview

Pulse is built for low-latency streaming, with multilingual, diarization and code-switching capabilities.

Pulse 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
Smallest
Hosted by
Smallest
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Multilingual, Diarization, Code switching

How Pulse 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 Pulse highlighted.
  1. #1STT RT v564 ms
  2. #2Parakeet TDT 0.6B v370 ms
  3. #3STT 183 ms
  4. #4Nova 399 ms
  5. #5Nova 2101 ms
  6. #6Ink 2108 ms
  7. #7Scribe v2 Realtime120 ms
  8. #8Universal 3.5 Pro146 ms
  9. #9Defaultvia Azure155 ms
  10. #10Whisper Large v3180 ms
  11. #11Velma 2 STT Streaming191 ms
  12. #12Grok STT199 ms
  13. #13Pulse205 ms
Show all 24 models
  1. #14Defaultvia Speechmatics223 ms
  2. #15Defaultvia Gradium249 ms
  3. #16resonant-1288 ms
  4. #17Enhanced341 ms
  5. #18Voxtral Mini Transcribe Realtime 2602356 ms
  6. #19GPT Realtime Whisper559 ms
  7. #20GPT-4o mini Transcribe623 ms
  8. #21Solaria 1680 ms
  9. #22GPT-4o Transcribe742 ms
  10. #23Chirp 2811 ms
  11. #24Chirp 3813 ms
median of all models · 202 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1STT RT v5Soniox64 ms1533 ms29,322
2Parakeet TDT 0.6B v3Together AI70 ms1210 ms29,161
3STT 1Inworld AI83 ms1468 ms29,280
4Nova 3Deepgram99 ms1434 ms29,290
5Nova 2Deepgram101 ms1436 ms29,145
6Ink 2Cartesia108 ms1812 ms29,288
7Scribe v2 RealtimeElevenLabs120 ms2157 ms29,313
8Universal 3.5 ProAssemblyAI146 ms1030 ms29,280
9DefaultAzure155 ms1792 ms6,682
10Whisper Large v3Together AI180 ms1241 ms29,171
11Velma 2 STT StreamingModulate191 ms1576 ms17,895
12Grok STTxAI199 ms—29,322
13PulseSmallest205 ms2072 ms29,313
14DefaultSpeechmatics223 ms1473 ms29,323
15DefaultGradium249 ms1979 ms28,235
16resonant-1Reson8288 ms—17,066
17EnhancedSpeechmatics341 ms1536 ms29,322
18Voxtral Mini Transcribe Realtime 2602Mistral356 ms1828 ms29,175
19GPT Realtime WhisperOpenAI559 ms1823 ms29,286
20GPT-4o mini TranscribeOpenAI623 ms—29,318
21Solaria 1Gladia680 ms1703 ms26,356
22GPT-4o TranscribeOpenAI742 ms—29,318
23Chirp 2Google811 ms6000 ms29,237
24Chirp 3Google813 ms5999 ms29,319
—FluxDeepgram—1089 ms29,354
—Flux MultilingualDeepgram—1175 ms29,344
—Nemotron 3.5 ASR StreamingTogether AI—1540 ms29,193
—Universal StreamingAssemblyAI—1510 ms29,297
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Highest relative placement: 12th of 28 on Word Error Rate.

Latency vs accuracy

TTFS vs WEREach point is one measured model · Pulse highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Pulse highlighted.
STT RT v5 — 64 ms, 5.8%Parakeet TDT 0.6B v3 — 70 ms, 11.0%STT 1 — 83 ms, 4.8%Nova 3 — 99 ms, 6.3%Nova 2 — 101 ms, 7.9%Ink 2 — 108 ms, 4.9%Scribe v2 Realtime — 120 ms, 5.5%Universal 3.5 Pro — 146 ms, 3.2%Default via Azure — 155 ms, 5.3%Whisper Large v3 — 180 ms, 8.2%Velma 2 STT Streaming — 191 ms, 6.4%Grok STT — 199 ms, 4.8%Default via Speechmatics — 223 ms, 5.5%Default via Gradium — 249 ms, 10.0%resonant-1 — 288 ms, 3.4%Enhanced — 341 ms, 4.3%Voxtral Mini Transcribe Realtime 2602 — 356 ms, 6.2%GPT Realtime Whisper — 559 ms, 5.2%GPT-4o mini Transcribe — 623 ms, 5.1%Solaria 1 — 680 ms, 7.1%GPT-4o Transcribe — 742 ms, 4.7%Chirp 2 — 811 ms, 5.0%Chirp 3 — 813 ms, 4.1%Pulse — 205 ms, 5.3%

Where the errors come from

WER compositionPulse's Word Error Rate split by error type · 30-day averagePulse's WER split into substitutions, deletions and insertions.
  • Pulse5.3%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Pulse latency distributionMilliseconds · shared axis across metrics · last 30 daysPulse's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to Final Segmentp50 193 ms · p99 373 ms
  • Time to First Tokenp50 2291 ms · p99 3518 ms
band p25–p75 · tick p50 · whisker to p99 with p90 and p95 stops
Average and percentile values per metric, with the number of samples behind each row.
MetricAveragep25p50p75p90p95p99Samples
Time to Final Segment205 ms185 ms193 ms206 ms254 ms293 ms373 ms29,313
Word Error Rate5.3%0.0%0.0%7.9%14.3%20.0%38.5%29,354
Time to First Token2072 ms1306 ms2291 ms2300 ms2323 ms2498 ms3518 ms29,354

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 daysPulse's daily median Time to Final Segment over the last 30 days.
Pulse · Jul 30: 191 msPulse · Jul 31: 190 msPulse · Aug 1: 191 msPulse · Aug 2: 189 msPulse · Aug 3: 193 msPulse · Aug 4: 190 msPulse · Aug 5: 191 msPulse · Aug 6: 195 msPulse · Aug 7: 199 msPulse · Aug 8: 188 msPulse · Aug 9: 196 msPulse · Aug 10: 195 msPulse · Aug 11: 187 msPulse · Aug 12: 187 msPulse · Aug 13: 190 msPulse · Aug 14: 197 msPulse · Aug 15: 189 msPulse · Aug 16: 196 msPulse · Aug 17: 193 msPulse · Aug 18: 194 msPulse · Aug 19: 195 msPulse · Aug 20: 195 msPulse · Aug 21: 195 msPulse · Aug 22: 193 msPulse · Aug 23: 186 msPulse · Aug 24: 196 msPulse · Aug 25: 202 msPulse · Aug 26: 197 msPulse · Aug 27: 194 msPulse · Aug 28: 201 ms
Word Error Rate — daily averageDaily average · UTC daysPulse's daily Word Error Rate over the last 30 days.
Pulse · Jul 30: 5.5%Pulse · Jul 31: 7.5%Pulse · Aug 1: 5.9%Pulse · Aug 2: 6.7%Pulse · Aug 3: 6.3%Pulse · Aug 4: 6.5%Pulse · Aug 5: 7.1%Pulse · Aug 6: 6.1%Pulse · Aug 7: 7.1%Pulse · Aug 8: 9.1%Pulse · Aug 9: 4.6%Pulse · Aug 10: 7.3%Pulse · Aug 11: 7.5%Pulse · Aug 12: 5.0%Pulse · Aug 13: 6.5%Pulse · Aug 14: 5.7%Pulse · Aug 15: 5.1%Pulse · Aug 16: 6.6%Pulse · Aug 17: 6.0%Pulse · Aug 18: 7.2%Pulse · Aug 19: 6.7%Pulse · Aug 20: 6.9%Pulse · Aug 21: 6.7%Pulse · Aug 22: 7.3%Pulse · Aug 23: 6.0%Pulse · Aug 24: 5.3%Pulse · Aug 25: 5.5%Pulse · Aug 26: 5.9%Pulse · Aug 27: 6.2%Pulse · Aug 28: 6.8%

Time to Final Segment by dataset

Pulse by test conditionMilliseconds · lower is better · best condition firstPulse's Time to Final Segment on each benchmark dataset.
  1. WildASR noise gaps202 ms
  2. PipeCat (production)202 ms
  3. WildASR clean205 ms
  4. WildASR accents209 ms
  5. WildASR clipping211 ms
  6. WildASR phone codec215 ms
  7. WildASR far-field215 ms
  8. WildASR reverb216 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
LibriSpeech193 ms1
ProductionPipeCat202 ms14,939
AccentsWildASR209 ms1,438
CleanWildASR205 ms5,780
ClippingWildASR211 ms1,444
Far-fieldWildASR215 ms1,442
Noise gapsWildASR202 ms1,441
Phone codecWildASR215 ms1,440
ReverbWildASR216 ms1,388

Strongest condition: LibriSpeech at 193 ms · weakest: WildASR reverb at 216 ms.

Limits of this comparison

Pulse is measured separately from the company's Lightning text-to-speech model.

  • The benchmark does not measure an end-to-end Pulse-plus-Lightning agent implementation.

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