Voice AI BenchmarksBy

SMALLESTSPEECH-TO-TEXT133,270 SAMPLES / 30 DAYSLAST RUN OCT 14, 2026, 16:30 UTC

official resource

Pulse speech-to-text benchmarks

Pulse, hosted by Smallest (Smallest, smallest.ai), measures mean 205 ms time to final segment (16th of 29) and 4.6% word error rate (17th of 31) among STT systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

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

Time to Final Segment#16 / 29
205ms
Word Error Rate#17 / 31
4.6%
Time to First Token#22 / 27
1976ms

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
Code switching, Diarization, Multilingual

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. #1Qwen3 ASR 1.7b39 ms
  2. #2STT RT v545 ms
  3. #3Qwen3 ASR Fast47 ms
  4. #4STT 167 ms
  5. #5Nova 386 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 Pro135 ms
  13. #13Ink 2149 ms
  14. #14Universal 3.5 Pro170 ms
  15. #15Whisper Large v3via Together AI197 ms
  16. #16Pulse205 ms
Show all 29 models
  1. #17Grok STT223 ms
  2. #18resonant-1252 ms
  3. #19Default262 ms
  4. #20Linden 1274 ms
  5. #21Gemini 3.5 Transcribe Live318 ms
  6. #22Daytona STT399 ms
  7. #23Voxtral Mini Transcribe Realtime 2602417 ms
  8. #24GPT Realtime Whisper599 ms
  9. #25Solaria 1619 ms
  10. #26GPT-4o mini Transcribe710 ms
  11. #27Chirp 3768 ms
  12. #28GPT-4o Transcribe784 ms
  13. #29Chirp 2909 ms
median of all models · 197 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Qwen3 ASR 1.7bBaseten39 ms946 ms13,669
2STT RT v5Soniox45 ms1518 ms23,533
3Qwen3 ASR FastNari Labs47 ms1752 ms26,652
4STT 1Inworld AI67 ms1205 ms24,748
5Nova 3Deepgram86 ms1432 ms26,002
6Nova 2Deepgram87 ms1417 ms25,960
7Parakeet TDT 0.6B v3Together AI97 ms1227 ms26,461
8Flux MultilingualDeepgram98 ms1165 ms26,007
9FluxDeepgram99 ms1091 ms26,007
10Scribe v2 RealtimeElevenLabs122 ms2157 ms26,640
11Whisper Large v3Baseten131 ms924 ms13,663
12Universal 3.6 ProAssemblyAI135 ms1071 ms24,603
13Ink 2Cartesia149 ms1969 ms26,653
14Universal 3.5 ProAssemblyAI170 ms1025 ms26,654
15Whisper Large v3Together AI197 ms1258 ms26,448
16PulseSmallest205 ms1976 ms26,654
17Grok STTSpaceXAI223 ms—26,577
18resonant-1Reson8252 ms—26,653
19DefaultGradium262 ms1987 ms26,632
20Linden 1Speechmatics274 ms1537 ms26,595
21Gemini 3.5 Transcribe LiveGemini318 ms1602 ms26,565
22Daytona STTGuava399 ms5695 ms5,225
23Voxtral Mini Transcribe Realtime 2602Mistral417 ms1853 ms26,041
24GPT Realtime WhisperOpenAI599 ms1824 ms26,622
25Solaria 1Gladia619 ms1766 ms26,360
26GPT-4o mini TranscribeOpenAI710 ms—26,621
27Chirp 3Google768 ms5956 ms26,638
28GPT-4o TranscribeOpenAI784 ms—26,625
29Chirp 2Google909 ms6099 ms26,650
—Nemotron 3.5 ASR StreamingTogether AI—1544 ms26,549
—Universal StreamingAssemblyAI—1509 ms26,626
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: 17th of 31 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.
Qwen3 ASR 1.7b — 39 ms, 2.9%STT RT v5 — 45 ms, 4.8%Qwen3 ASR Fast — 47 ms, 2.6%STT 1 — 67 ms, 7.1%Nova 3 — 86 ms, 5.3%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 — 135 ms, 2.1%Ink 2 — 149 ms, 3.6%Universal 3.5 Pro — 170 ms, 2.4%Whisper Large v3 via Together AI — 197 ms, 7.4%Grok STT — 223 ms, 4.1%resonant-1 — 252 ms, 2.6%Default — 262 ms, 8.8%Linden 1 — 274 ms, 3.9%Gemini 3.5 Transcribe Live — 318 ms, 3.0%Daytona STT — 399 ms, 4.9%Voxtral Mini Transcribe Realtime 2602 — 417 ms, 5.3%GPT Realtime Whisper — 599 ms, 4.0%Solaria 1 — 619 ms, 6.1%GPT-4o mini Transcribe — 710 ms, 3.9%Chirp 3 — 768 ms, 3.2%GPT-4o Transcribe — 784 ms, 3.8%Chirp 2 — 909 ms, 3.9%Pulse — 205 ms, 4.6%

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.
  • Pulse4.6%
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 max.
  • Time to Final Segmentp50 192 ms · Max 1172 ms
  • Time to First Tokenp50 2087 ms · Max 5555 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 Segment205 ms185 ms192 ms208 ms253 ms290 ms1172 ms26,654
Word Error Rate4.6%0.0%2.6%7.7%13.6%19.2%100.0%26,654
Time to First Token1976 ms2076 ms2087 ms2096 ms2141 ms3283 ms5555 ms26,654

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 · Sep 14: 199 msPulse · Sep 15: 194 msPulse · Sep 16: 195 msPulse · Sep 17: 187 msPulse · Sep 18: 194 msPulse · Sep 19: 194 msPulse · Sep 20: 189 msPulse · Sep 21: 195 msPulse · Sep 22: 203 msPulse · Sep 23: 203 msPulse · Sep 24: 192 msPulse · Sep 25: 194 msPulse · Sep 26: 196 msPulse · Sep 27: 190 msPulse · Sep 28: 246 msPulse · Sep 29: 196 msPulse · Sep 30: 188 msPulse · Oct 1: 191 msPulse · Oct 2: 193 msPulse · Oct 3: 190 msPulse · Oct 4: 191 msPulse · Oct 5: 190 msPulse · Oct 6: 189 msPulse · Oct 7: 191 msPulse · Oct 8: 194 msPulse · Oct 9: 192 msPulse · Oct 10: 195 msPulse · Oct 11: 194 msPulse · Oct 12: 193 msPulse · Oct 13: 190 msPulse · Oct 14: 192 ms
Word Error Rate — daily averageDaily average · UTC daysPulse's daily Word Error Rate over the last 30 days.
Pulse · Sep 14: 5.2%Pulse · Sep 15: 5.4%Pulse · Sep 16: 5.6%Pulse · Sep 17: 6.4%Pulse · Sep 18: 6.1%Pulse · Sep 19: 4.4%Pulse · Sep 20: 6.1%Pulse · Sep 21: 5.6%Pulse · Sep 22: 6.5%Pulse · Sep 23: 5.2%Pulse · Sep 24: 5.9%Pulse · Sep 25: 5.9%Pulse · Sep 26: 5.8%Pulse · Sep 27: 5.3%Pulse · Sep 28: 5.5%Pulse · Sep 29: 4.7%Pulse · Sep 30: 6.0%Pulse · Oct 1: 5.0%Pulse · Oct 2: 3.9%Pulse · Oct 3: 5.8%Pulse · Oct 4: 6.9%Pulse · Oct 5: 5.2%Pulse · Oct 6: 6.3%Pulse · Oct 7: 4.9%Pulse · Oct 8: 5.6%Pulse · Oct 9: 4.8%Pulse · Oct 10: 5.1%Pulse · Oct 11: 5.7%Pulse · Oct 12: 5.7%Pulse · Oct 13: 4.9%Pulse · Oct 14: 5.4%

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. PipeCat (production)202 ms
  2. WildASR noise gaps205 ms
  3. WildASR clean206 ms
  4. WildASR accents210 ms
  5. WildASR clipping211 ms
  6. WildASR far-field214 ms
  7. WildASR phone codec218 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
ProductionPipeCat202 ms14,020
AccentsWildASR210 ms1,403
CleanWildASR206 ms5,612
ClippingWildASR211 ms1,405
Far-fieldWildASR214 ms1,406
Noise gapsWildASR205 ms1,403
Phone codecWildASR218 ms1,405

Strongest condition: PipeCat (production) at 202 ms · weakest: WildASR phone codec at 218 ms.

How fast is Pulse?

On Smallest, Pulse measures mean 205 ms time to final segment (16th of 29) and mean 1976 ms time to first token (22nd of 27). Last measured 2026-10-14.

How accurate is Pulse?

On Smallest, Pulse measures 4.6% word error rate (17th of 31). Last measured 2026-10-14.

Who hosts Pulse?

Pulse is created by Smallest and served by Smallest. Coval measures each hosted endpoint separately.

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