Voice AI BenchmarksBy

ASSEMBLYAISPEECH-TO-TEXT146,542 SAMPLES / 30 DAYSLAST RUN AUG 28, 2026, 21:30 UTC

official resource

Universal 3.5 Pro speech-to-text benchmarks

Universal 3.5 Pro is AssemblyAI's accuracy-oriented Universal recognizer.

Time to Final Segment#8 / 24
146ms
Word Error Rate#1 / 28
3.2%
Time to First Token#1 / 24
1030ms

Overview

The tested AssemblyAI endpoint supports multilingual recognition, diarization, code switching and keyterm controls.

Universal 3.5 Pro 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
AssemblyAI
Hosted by
AssemblyAI
Source
Official API
Licensing
Proprietary
Deployment
On-prem
Region
US
Features
Multilingual, VAD, Diarization, Code switching, Keyterm biasing

How Universal 3.5 Pro 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 Universal 3.5 Pro 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
Show all 24 models
  1. #13Pulse205 ms
  2. #14Defaultvia Speechmatics223 ms
  3. #15Defaultvia Gradium249 ms
  4. #16resonant-1288 ms
  5. #17Enhanced341 ms
  6. #18Voxtral Mini Transcribe Realtime 2602356 ms
  7. #19GPT Realtime Whisper559 ms
  8. #20GPT-4o mini Transcribe623 ms
  9. #21Solaria 1680 ms
  10. #22GPT-4o Transcribe742 ms
  11. #23Chirp 2811 ms
  12. #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: 1st of 28 on Word Error Rate.

Latency vs accuracy

TTFS vs WEREach point is one measured model · Universal 3.5 Pro highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Universal 3.5 Pro 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%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%Pulse — 205 ms, 5.3%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%Universal 3.5 Pro — 146 ms, 3.2%

Where the errors come from

WER compositionUniversal 3.5 Pro's Word Error Rate split by error type · 30-day averageUniversal 3.5 Pro's WER split into substitutions, deletions and insertions.
  • Universal 3.5 Pro3.2%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Universal 3.5 Pro latency distributionMilliseconds · shared axis across metrics · last 30 daysUniversal 3.5 Pro's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to Final Segmentp50 118 ms · p99 458 ms
  • Time to First Tokenp50 1117 ms · p99 2918 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 Segment146 ms90 ms118 ms166 ms228 ms288 ms458 ms29,280
Word Error Rate3.2%0.0%0.0%3.6%9.4%14.3%33.3%29,315
Time to First Token1030 ms622 ms1117 ms1217 ms1518 ms1818 ms2918 ms29,317

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 daysUniversal 3.5 Pro's daily median Time to Final Segment over the last 30 days.
Universal 3.5 Pro · Jul 30: 128 msUniversal 3.5 Pro · Jul 31: 98 msUniversal 3.5 Pro · Aug 1: 83 msUniversal 3.5 Pro · Aug 2: 84 msUniversal 3.5 Pro · Aug 3: 99 msUniversal 3.5 Pro · Aug 4: 110 msUniversal 3.5 Pro · Aug 5: 105 msUniversal 3.5 Pro · Aug 6: 104 msUniversal 3.5 Pro · Aug 7: 124 msUniversal 3.5 Pro · Aug 8: 107 msUniversal 3.5 Pro · Aug 9: 132 msUniversal 3.5 Pro · Aug 10: 120 msUniversal 3.5 Pro · Aug 11: 130 msUniversal 3.5 Pro · Aug 12: 148 msUniversal 3.5 Pro · Aug 13: 120 msUniversal 3.5 Pro · Aug 14: 121 msUniversal 3.5 Pro · Aug 15: 103 msUniversal 3.5 Pro · Aug 16: 108 msUniversal 3.5 Pro · Aug 17: 117 msUniversal 3.5 Pro · Aug 18: 116 msUniversal 3.5 Pro · Aug 19: 143 msUniversal 3.5 Pro · Aug 20: 122 msUniversal 3.5 Pro · Aug 21: 127 msUniversal 3.5 Pro · Aug 22: 132 msUniversal 3.5 Pro · Aug 23: 118 msUniversal 3.5 Pro · Aug 24: 144 msUniversal 3.5 Pro · Aug 25: 147 msUniversal 3.5 Pro · Aug 26: 152 msUniversal 3.5 Pro · Aug 27: 159 msUniversal 3.5 Pro · Aug 28: 150 ms
Word Error Rate — daily averageDaily average · UTC daysUniversal 3.5 Pro's daily Word Error Rate over the last 30 days.
Universal 3.5 Pro · Jul 30: 5.3%Universal 3.5 Pro · Jul 31: 4.9%Universal 3.5 Pro · Aug 1: 3.4%Universal 3.5 Pro · Aug 2: 4.9%Universal 3.5 Pro · Aug 3: 2.7%Universal 3.5 Pro · Aug 4: 4.3%Universal 3.5 Pro · Aug 5: 5.0%Universal 3.5 Pro · Aug 6: 2.9%Universal 3.5 Pro · Aug 7: 5.3%Universal 3.5 Pro · Aug 8: 5.4%Universal 3.5 Pro · Aug 9: 5.2%Universal 3.5 Pro · Aug 10: 4.1%Universal 3.5 Pro · Aug 11: 4.5%Universal 3.5 Pro · Aug 12: 3.5%Universal 3.5 Pro · Aug 13: 3.5%Universal 3.5 Pro · Aug 14: 4.0%Universal 3.5 Pro · Aug 15: 3.3%Universal 3.5 Pro · Aug 16: 4.1%Universal 3.5 Pro · Aug 17: 3.5%Universal 3.5 Pro · Aug 18: 4.4%Universal 3.5 Pro · Aug 19: 4.7%Universal 3.5 Pro · Aug 20: 3.7%Universal 3.5 Pro · Aug 21: 4.1%Universal 3.5 Pro · Aug 22: 5.8%Universal 3.5 Pro · Aug 23: 4.7%Universal 3.5 Pro · Aug 24: 3.5%Universal 3.5 Pro · Aug 25: 4.3%Universal 3.5 Pro · Aug 26: 3.4%Universal 3.5 Pro · Aug 27: 4.2%Universal 3.5 Pro · Aug 28: 4.7%

Time to Final Segment by dataset

Universal 3.5 Pro by test conditionMilliseconds · lower is better · best condition firstUniversal 3.5 Pro's Time to Final Segment on each benchmark dataset.
  1. WildASR clean140 ms
  2. WildASR phone codec143 ms
  3. PipeCat (production)145 ms
  4. WildASR noise gaps145 ms
  5. WildASR clipping147 ms
  6. WildASR reverb147 ms
  7. WildASR far-field151 ms
  8. WildASR accents173 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
LibriSpeech222 ms1
ProductionPipeCat145 ms14,919
AccentsWildASR173 ms1,438
CleanWildASR140 ms5,769
ClippingWildASR147 ms1,441
Far-fieldWildASR151 ms1,438
Noise gapsWildASR145 ms1,438
Phone codecWildASR143 ms1,440
ReverbWildASR147 ms1,396

Strongest condition: WildASR clean at 140 ms · weakest: LibriSpeech at 222 ms.

Limits of this comparison

Coval keeps it independent from Universal Streaming so different product modes are not collapsed into one brand result.

  • Timing reflects Coval's API mode and should not be assumed to describe every asynchronous transcription workflow.

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