Voice AI BenchmarksBy

OPENAISPEECH-TO-TEXT117,395 SAMPLES / 30 DAYSLAST RUN AUG 28, 2026, 21:30 UTC

official resource

GPT-4o Transcribe speech-to-text benchmarks

GPT-4o Transcribe is OpenAI's GPT-4o-based speech recognition model for the Audio API.

Time to Final Segment#22 / 24
742ms
Word Error Rate#5 / 28
4.7%

Overview

GPT-4o Transcribe is a dedicated transcription model rather than the transcription path inside an OpenAI Realtime session.

GPT-4o Transcribe 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
OpenAI
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Multilingual, VAD, Keyterm biasing

How GPT-4o Transcribe 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 GPT-4o Transcribe 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
  14. #14Defaultvia Speechmatics223 ms
  15. #15Defaultvia Gradium249 ms
  16. #16resonant-1288 ms
  17. #17Enhanced341 ms
  18. #18Voxtral Mini Transcribe Realtime 2602356 ms
  19. #19GPT Realtime Whisper559 ms
  20. #20GPT-4o mini Transcribe623 ms
  21. #21Solaria 1680 ms
  22. #22GPT-4o Transcribe742 ms
Show all 24 models
  1. #23Chirp 2811 ms
  2. #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 28 on Word Error Rate.

Latency vs accuracy

TTFS vs WEREach point is one measured model · GPT-4o Transcribe highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with GPT-4o Transcribe 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%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%Chirp 2 — 811 ms, 5.0%Chirp 3 — 813 ms, 4.1%GPT-4o Transcribe — 742 ms, 4.7%

Where the errors come from

WER compositionGPT-4o Transcribe's Word Error Rate split by error type · 30-day averageGPT-4o Transcribe's WER split into substitutions, deletions and insertions.
  • GPT-4o Transcribe4.7%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

GPT-4o Transcribe latency distributionMilliseconds · shared axis across metrics · last 30 daysGPT-4o Transcribe's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to Final Segmentp50 714 ms · p99 1441 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 Segment742 ms619 ms714 ms815 ms924 ms1029 ms1441 ms29,318
Word Error Rate4.7%0.0%0.0%5.3%13.3%21.4%53.3%29,359

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 daysGPT-4o Transcribe's daily median Time to Final Segment over the last 30 days.
GPT-4o Transcribe · Jul 30: 786 msGPT-4o Transcribe · Jul 31: 680 msGPT-4o Transcribe · Aug 1: 740 msGPT-4o Transcribe · Aug 2: 654 msGPT-4o Transcribe · Aug 3: 798 msGPT-4o Transcribe · Aug 4: 725 msGPT-4o Transcribe · Aug 5: 735 msGPT-4o Transcribe · Aug 6: 704 msGPT-4o Transcribe · Aug 7: 738 msGPT-4o Transcribe · Aug 8: 699 msGPT-4o Transcribe · Aug 9: 695 msGPT-4o Transcribe · Aug 10: 754 msGPT-4o Transcribe · Aug 11: 672 msGPT-4o Transcribe · Aug 12: 772 msGPT-4o Transcribe · Aug 13: 718 msGPT-4o Transcribe · Aug 14: 650 msGPT-4o Transcribe · Aug 15: 700 msGPT-4o Transcribe · Aug 16: 676 msGPT-4o Transcribe · Aug 17: 709 msGPT-4o Transcribe · Aug 18: 727 msGPT-4o Transcribe · Aug 19: 722 msGPT-4o Transcribe · Aug 20: 670 msGPT-4o Transcribe · Aug 21: 756 msGPT-4o Transcribe · Aug 22: 699 msGPT-4o Transcribe · Aug 23: 641 msGPT-4o Transcribe · Aug 24: 690 msGPT-4o Transcribe · Aug 25: 781 msGPT-4o Transcribe · Aug 26: 786 msGPT-4o Transcribe · Aug 27: 660 msGPT-4o Transcribe · Aug 28: 759 ms
Word Error Rate — daily averageDaily average · UTC daysGPT-4o Transcribe's daily Word Error Rate over the last 30 days.
GPT-4o Transcribe · Jul 30: 4.4%GPT-4o Transcribe · Jul 31: 6.0%GPT-4o Transcribe · Aug 1: 5.3%GPT-4o Transcribe · Aug 2: 6.9%GPT-4o Transcribe · Aug 3: 4.4%GPT-4o Transcribe · Aug 4: 7.2%GPT-4o Transcribe · Aug 5: 6.5%GPT-4o Transcribe · Aug 6: 4.8%GPT-4o Transcribe · Aug 7: 6.9%GPT-4o Transcribe · Aug 8: 5.0%GPT-4o Transcribe · Aug 9: 6.3%GPT-4o Transcribe · Aug 10: 6.8%GPT-4o Transcribe · Aug 11: 6.1%GPT-4o Transcribe · Aug 12: 5.3%GPT-4o Transcribe · Aug 13: 4.9%GPT-4o Transcribe · Aug 14: 7.0%GPT-4o Transcribe · Aug 15: 5.7%GPT-4o Transcribe · Aug 16: 6.3%GPT-4o Transcribe · Aug 17: 4.5%GPT-4o Transcribe · Aug 18: 6.7%GPT-4o Transcribe · Aug 19: 6.5%GPT-4o Transcribe · Aug 20: 6.1%GPT-4o Transcribe · Aug 21: 4.3%GPT-4o Transcribe · Aug 22: 6.5%GPT-4o Transcribe · Aug 23: 5.2%GPT-4o Transcribe · Aug 24: 5.1%GPT-4o Transcribe · Aug 25: 5.5%GPT-4o Transcribe · Aug 26: 5.5%GPT-4o Transcribe · Aug 27: 5.1%GPT-4o Transcribe · Aug 28: 6.6%

Time to Final Segment by dataset

GPT-4o Transcribe by test conditionMilliseconds · lower is better · best condition firstGPT-4o Transcribe's Time to Final Segment on each benchmark dataset.
  1. WildASR accents638 ms
  2. WildASR reverb737 ms
  3. WildASR far-field741 ms
  4. WildASR clipping743 ms
  5. WildASR phone codec743 ms
  6. WildASR clean744 ms
  7. WildASR noise gaps750 ms
  8. PipeCat (production)751 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
LibriSpeech1028 ms1
ProductionPipeCat751 ms14,934
AccentsWildASR638 ms1,438
CleanWildASR744 ms5,779
ClippingWildASR743 ms1,444
Far-fieldWildASR741 ms1,442
Noise gapsWildASR750 ms1,441
Phone codecWildASR743 ms1,441
ReverbWildASR737 ms1,398

Strongest condition: WildASR accents at 638 ms · weakest: LibriSpeech at 1028 ms.

Limits of this comparison

Coval reports accuracy, first partial output and final-segment timing separately because a fast partial does not guarantee a fast settled transcript.

  • Prompting, language mix and audio format can affect production behavior beyond Coval's fixed public datasets.

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