Voice AI BenchmarksBy

OPENAISPEECH-TO-TEXT145,960 SAMPLES / 30 DAYSLAST RUN AUG 28, 2026, 21:30 UTC

official resource

Whisper Large v3 speech-to-text benchmarks

Whisper Large v3 is OpenAI's downloadable multilingual recognizer.

Time to Final Segment#10 / 24
180ms
Word Error Rate#24 / 28
8.2%
Time to First Token#5 / 24
1241ms

Overview

OpenAI publishes Whisper's code and model weights, while Together supplies the measured inference endpoint.

Whisper Large v3 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
OpenAI
Hosted by
Together AI
Source
Shared inference
Licensing
Open-weight
Deployment
Cloud
Region
US
Features
Multilingual, VAD

How Whisper Large v3 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 Whisper Large v3 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: 5th of 24 on Time to First Token.

Latency vs accuracy

TTFS vs WEREach point is one measured model · Whisper Large v3 highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Whisper Large v3 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%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%Whisper Large v3 — 180 ms, 8.2%

Where the errors come from

WER compositionWhisper Large v3's Word Error Rate split by error type · 30-day averageWhisper Large v3's WER split into substitutions, deletions and insertions.
  • Whisper Large v38.2%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Whisper Large v3 latency distributionMilliseconds · shared axis across metrics · last 30 daysWhisper Large v3's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to Final Segmentp50 112 ms · p99 1641 ms
  • Time to First Tokenp50 1320 ms · p99 3480 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 Segment180 ms91 ms112 ms149 ms214 ms448 ms1641 ms29,171
Word Error Rate8.2%0.0%5.0%11.8%20.0%27.3%53.8%29,232
Time to First Token1241 ms840 ms1320 ms1446 ms1710 ms2097 ms3480 ms29,133

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 daysWhisper Large v3's daily median Time to Final Segment over the last 30 days.
Whisper Large v3 · Jul 30: 93 msWhisper Large v3 · Jul 31: 100 msWhisper Large v3 · Aug 1: 106 msWhisper Large v3 · Aug 2: 112 msWhisper Large v3 · Aug 3: 103 msWhisper Large v3 · Aug 4: 102 msWhisper Large v3 · Aug 5: 106 msWhisper Large v3 · Aug 6: 106 msWhisper Large v3 · Aug 7: 104 msWhisper Large v3 · Aug 8: 95 msWhisper Large v3 · Aug 9: 109 msWhisper Large v3 · Aug 10: 96 msWhisper Large v3 · Aug 11: 111 msWhisper Large v3 · Aug 12: 121 msWhisper Large v3 · Aug 13: 130 msWhisper Large v3 · Aug 14: 102 msWhisper Large v3 · Aug 15: 119 msWhisper Large v3 · Aug 16: 98 msWhisper Large v3 · Aug 17: 123 msWhisper Large v3 · Aug 18: 127 msWhisper Large v3 · Aug 19: 103 msWhisper Large v3 · Aug 20: 101 msWhisper Large v3 · Aug 21: 123 msWhisper Large v3 · Aug 22: 130 msWhisper Large v3 · Aug 23: 111 msWhisper Large v3 · Aug 24: 137 msWhisper Large v3 · Aug 25: 160 msWhisper Large v3 · Aug 26: 128 msWhisper Large v3 · Aug 27: 162 msWhisper Large v3 · Aug 28: 177 ms
Word Error Rate — daily averageDaily average · UTC daysWhisper Large v3's daily Word Error Rate over the last 30 days.
Whisper Large v3 · Jul 30: 8.9%Whisper Large v3 · Jul 31: 8.8%Whisper Large v3 · Aug 1: 8.6%Whisper Large v3 · Aug 2: 11.3%Whisper Large v3 · Aug 3: 7.4%Whisper Large v3 · Aug 4: 9.0%Whisper Large v3 · Aug 5: 11.2%Whisper Large v3 · Aug 6: 6.7%Whisper Large v3 · Aug 7: 8.6%Whisper Large v3 · Aug 8: 9.9%Whisper Large v3 · Aug 9: 7.5%Whisper Large v3 · Aug 10: 9.8%Whisper Large v3 · Aug 11: 8.5%Whisper Large v3 · Aug 12: 9.4%Whisper Large v3 · Aug 13: 8.8%Whisper Large v3 · Aug 14: 9.3%Whisper Large v3 · Aug 15: 7.8%Whisper Large v3 · Aug 16: 9.3%Whisper Large v3 · Aug 17: 7.7%Whisper Large v3 · Aug 18: 8.8%Whisper Large v3 · Aug 19: 8.7%Whisper Large v3 · Aug 20: 8.3%Whisper Large v3 · Aug 21: 7.5%Whisper Large v3 · Aug 22: 10.8%Whisper Large v3 · Aug 23: 7.7%Whisper Large v3 · Aug 24: 10.8%Whisper Large v3 · Aug 25: 9.4%Whisper Large v3 · Aug 26: 8.5%Whisper Large v3 · Aug 27: 10.3%Whisper Large v3 · Aug 28: 9.5%

Time to Final Segment by dataset

Whisper Large v3 by test conditionMilliseconds · lower is better · best condition firstWhisper Large v3's Time to Final Segment on each benchmark dataset.
  1. WildASR noise gaps156 ms
  2. WildASR far-field166 ms
  3. WildASR clipping166 ms
  4. WildASR reverb176 ms
  5. WildASR accents177 ms
  6. WildASR clean178 ms
  7. WildASR phone codec185 ms
  8. PipeCat (production)186 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
LibriSpeech128 ms1
ProductionPipeCat186 ms14,865
AccentsWildASR177 ms1,427
CleanWildASR178 ms5,751
ClippingWildASR166 ms1,436
Far-fieldWildASR166 ms1,434
Noise gapsWildASR156 ms1,439
Phone codecWildASR185 ms1,432
ReverbWildASR176 ms1,386

Strongest condition: LibriSpeech at 128 ms · weakest: PipeCat (production) at 186 ms.

Limits of this comparison

Coval measures it through Together-hosted inference, so the latency results apply to Together's serving configuration.

  • Quantization, hardware, batching and decoding parameters can change speed and accuracy for an open-weight model.

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