ZOOMSPEECH-TO-TEXT85,368 SAMPLES / 30 DAYSLAST RUN NOV 14, 2026, 01:30 UTC

Scribe speech-to-text benchmarks

Scribe, hosted by Zoom, measures mean 112 ms time to final segment (15th of 30) and 2.6% word error rate (8th of 31) among STT systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

Coval has active speech-to-text measurements for Scribe, created by Zoom and served through Zoom's own API.

Word Error Rate#8 / 31
2.6%

Overview

Scribe 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
Zoom
Hosted by
Zoom
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Multilingual, VAD

How Scribe 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 Scribe highlighted.
  1. #1Wire-129 ms
  2. #2Qwen3 ASR 1.7b35 ms
  3. #3STT RT v551 ms
  4. #5STT Rt v659 ms
  5. #6STT 170 ms
  6. #7Nova 395 ms
  7. #8Nova 297 ms
  8. #9Rumble-298 ms
  9. #14Flux112 ms
  10. #15Scribe112 ms
Show all 30 models
  1. #17Whisper Large v3125 ms
  2. #18Ink 2141 ms
  3. #19Grok STT163 ms
  4. #20Pulse168 ms
  5. #21resonant-1210 ms
  6. #22Linden 1229 ms
  7. #24Default284 ms
  8. #25Daytona STT347 ms
  9. #27Solaria 1390 ms
  10. #29Chirp 3842 ms
  11. #30Chirp 2909 ms
median of all models · 114 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Saidwire En 2026 10 30Saidwire29 ms768 ms9,460
2Qwen3 ASR 1.7bBaseten35 ms941 ms13,535
3STT RT v5Soniox51 ms1529 ms27,360
4Qwen3 ASR FastNari Labs53 ms1757 ms27,363
5STT Rt v6Soniox59 ms1893 ms12,207
6STT 1Inworld AI70 ms994 ms27,336
7Nova 3Deepgram95 ms1435 ms24,082
8Nova 2Deepgram97 ms1432 ms24,047
9Realtime GlobalReson898 ms1658 ms27,363
10Scribe v3 LiveElevenLabs98 ms1254 ms5,800
11Universal 3.5 ProAssemblyAI104 ms1013 ms27,363
12Universal 3.6 ProAssemblyAI110 ms1073 ms27,362
13Flux MultilingualDeepgram110 ms1173 ms24,080
14FluxDeepgram112 ms1095 ms24,086
15ScribeZoom112 ms—6,654
16Scribe v2 RealtimeElevenLabs116 ms2126 ms27,359
17Whisper Large v3Baseten125 ms919 ms13,079
18Ink 2Cartesia141 ms2030 ms18,414
19Grok STTSpaceXAI163 ms—27,345
20PulseSmallest168 ms1953 ms27,360
21resonant-1Reson8210 ms—27,355
22Linden 1Speechmatics229 ms1503 ms27,339
23Scribe Pro v2.1Zoom241 ms—6,823
24DefaultGradium284 ms2074 ms27,355
25Daytona STTGuava347 ms5606 ms13,249
26Gemini 3.5 Transcribe LiveGemini355 ms1520 ms27,283
27Solaria 1Gladia390 ms1535 ms27,320
28GPT Realtime WhisperOpenAI566 ms1813 ms26,880
29Chirp 3Google842 ms6035 ms27,313
30Chirp 2Google909 ms6101 ms27,361
—Universal StreamingAssemblyAI—1511 ms27,332
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: 8th of 31 on Word Error Rate.

Latency vs accuracy

Where the errors come from

WER compositionScribe's Word Error Rate split by error type · 30-day averageScribe's WER split into substitutions, deletions and insertions.
  • Scribe2.6%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Scribe latency distributionMilliseconds · shared axis across metrics · last 30 daysScribe's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to Final Segmentp50 110 ms · Max 363 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 Segment112 ms73 ms110 ms150 ms174 ms187 ms363 ms6,654
Word Error Rate2.6%0.0%0.0%3.4%9.5%13.6%66.7%24,543

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 daysScribe's daily median Time to Final Segment over the last 30 days.
Scribe · Nov 6: 117 msScribe · Nov 7: 110 msScribe · Nov 8: 100 msScribe · Nov 9: 105 msScribe · Nov 10: 110 msScribe · Nov 11: 113 msScribe · Nov 12: 110 msScribe · Nov 13: 109 msScribe · Nov 14: 120 ms
Word Error Rate — dailyDaily corpus-level · UTC daysScribe's daily Word Error Rate over the last 30 days.
Scribe · Oct 18: 3.4%Scribe · Oct 19: 3.3%Scribe · Oct 20: 3.6%Scribe · Oct 21: 3.7%Scribe · Oct 22: 3.3%Scribe · Oct 23: 3.3%Scribe · Oct 24: 3.4%Scribe · Oct 25: 3.2%Scribe · Oct 26: 3.8%Scribe · Oct 27: 4.5%Scribe · Oct 28: 3.3%Scribe · Oct 29: 3.0%Scribe · Oct 30: 3.4%Scribe · Oct 31: 3.4%Scribe · Nov 1: 3.0%Scribe · Nov 2: 3.9%Scribe · Nov 3: 3.5%Scribe · Nov 4: 3.7%Scribe · Nov 5: 3.8%Scribe · Nov 6: 4.2%Scribe · Nov 7: 2.6%Scribe · Nov 8: 3.2%Scribe · Nov 9: 3.2%Scribe · Nov 10: 2.0%Scribe · Nov 11: 1.6%Scribe · Nov 12: 1.7%Scribe · Nov 13: 1.6%Scribe · Nov 14: 0.5%

Time to Final Segment by dataset

Scribe by test conditionMilliseconds · lower is better · best condition firstScribe's Time to Final Segment on each benchmark dataset.
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
ProductionPipeCat116 ms3,499
AccentsWildASR169 ms351
CleanWildASR99 ms1,404
ClippingWildASR101 ms349
Far-fieldWildASR102 ms351
Noise gapsWildASR103 ms351
Phone codecWildASR104 ms349

Strongest condition: WildASR clean at 99 ms · weakest: WildASR accents at 169 ms.

How fast is Scribe?

On Zoom, Scribe measures mean 112 ms time to final segment (15th of 30). Last measured 2026-11-14.

How accurate is Scribe?

On Zoom, Scribe measures 2.6% word error rate (8th of 31). Last measured 2026-11-14.

Who hosts Scribe?

Scribe is created by Zoom and served by Zoom. Coval measures each hosted endpoint separately.

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