Voice AI BenchmarksBy

SONIOXSPEECH-TO-TEXT7,305 SAMPLES / 30 DAYSLAST RUN NOV 5, 2026, 08:30 UTC

STT Rt v6 speech-to-text benchmarks

STT Rt v6, hosted by Soniox, measures mean 57 ms time to final segment (4th of 28) and 3.9% word error rate (11th of 30) 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 STT Rt v6, created by Soniox and served through Soniox's own API.

Time to Final Segment#4 / 28
57ms
Word Error Rate#11 / 30
3.9%
Time to First Token#20 / 28
1888ms

Overview

STT Rt v6 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
Soniox
Hosted by
Soniox
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Code switching, Diarization, Keyterm biasing, Multilingual, Translation, VAD

How STT Rt v6 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 STT Rt v6 highlighted.
  1. #1Qwen3 ASR 1.7b35 ms
  2. #2STT RT v546 ms
  3. #3Qwen3 ASR Fast51 ms
  4. #4STT Rt v657 ms
  5. #5STT 170 ms
  6. #6Nova 386 ms
  7. #7Nova 289 ms
  8. #8Flux Multilingual101 ms
  9. #9Flux102 ms
  10. #10Universal 3.6 Pro110 ms
  11. #11Scribe v2 Realtime119 ms
  12. #12Universal 3.5 Pro119 ms
Show all 28 models
  1. #13Whisper Large v3via Baseten127 ms
  2. #14Ink 2145 ms
  3. #15Rumble-2147 ms
  4. #16Whisper Large v3via Together AI173 ms
  5. #17Grok STT181 ms
  6. #18Pulse203 ms
  7. #19resonant-1237 ms
  8. #20Linden 1239 ms
  9. #21Default266 ms
  10. #22Daytona STT375 ms
  11. #23Gemini 3.5 Transcribe Live381 ms
  12. #24Solaria 1416 ms
  13. #25Voxtral Mini Transcribe Realtime 2602444 ms
  14. #26GPT Realtime Whisper564 ms
  15. #27Chirp 3832 ms
  16. #28Chirp 2897 ms
median of all models · 146 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Qwen3 ASR 1.7bBaseten35 ms939 ms13,679
2STT RT v5Soniox46 ms1526 ms27,363
3Qwen3 ASR FastNari Labs51 ms1755 ms27,370
4STT Rt v6Soniox57 ms1888 ms1,461
5STT 1Inworld AI70 ms1010 ms27,342
6Nova 3Deepgram86 ms1439 ms27,362
7Nova 2Deepgram89 ms1427 ms27,317
8Flux MultilingualDeepgram101 ms1173 ms27,363
9FluxDeepgram102 ms1095 ms27,369
10Universal 3.6 ProAssemblyAI110 ms1077 ms27,332
11Scribe v2 RealtimeElevenLabs119 ms2150 ms27,366
12Universal 3.5 ProAssemblyAI119 ms1016 ms27,371
13Whisper Large v3Baseten127 ms917 ms13,678
14Ink 2Cartesia145 ms2033 ms27,368
15Realtime GlobalReson8147 ms1977 ms25,204
16Whisper Large v3Together AI173 ms1197 ms12,785
17Grok STTSpaceXAI181 ms—27,337
18PulseSmallest203 ms1988 ms27,365
19resonant-1Reson8237 ms—27,371
20Linden 1Speechmatics239 ms1511 ms27,348
21DefaultGradium266 ms2015 ms27,365
22Daytona STTGuava375 ms5648 ms13,194
23Gemini 3.5 Transcribe LiveGemini381 ms1533 ms27,270
24Solaria 1Gladia416 ms1606 ms27,288
25Voxtral Mini Transcribe Realtime 2602Mistral444 ms1872 ms20,725
26GPT Realtime WhisperOpenAI564 ms1820 ms26,860
27Chirp 3Google832 ms6030 ms27,319
28Chirp 2Google897 ms6095 ms27,367
—Nemotron 3.5 ASR StreamingTogether AI—1562 ms12,697
—Universal StreamingAssemblyAI—1514 ms27,336
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Latency vs accuracy

TTFS vs WEREach point is one measured model · STT Rt v6 highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with STT Rt v6 highlighted.
Qwen3 ASR 1.7b — 35 ms, 2.9%STT RT v5 — 46 ms, 4.7%Qwen3 ASR Fast — 51 ms, 2.6%STT 1 — 70 ms, 4.5%Nova 3 — 86 ms, 5.2%Nova 2 — 89 ms, 6.6%Flux Multilingual — 101 ms, 7.5%Flux — 102 ms, 5.7%Universal 3.6 Pro — 110 ms, 2.2%Scribe v2 Realtime — 119 ms, 4.2%Universal 3.5 Pro — 119 ms, 2.4%Whisper Large v3 via Baseten — 127 ms, 4.5%Ink 2 — 145 ms, 3.4%Rumble-2 — 147 ms, 2.6%Whisper Large v3 via Together AI — 173 ms, 7.2%Grok STT — 181 ms, 4.1%Pulse — 203 ms, 4.6%resonant-1 — 237 ms, 2.6%Linden 1 — 239 ms, 3.8%Default — 266 ms, 8.2%Daytona STT — 375 ms, 4.8%Gemini 3.5 Transcribe Live — 381 ms, 2.9%Solaria 1 — 416 ms, 6.2%Voxtral Mini Transcribe Realtime 2602 — 444 ms, 5.4%GPT Realtime Whisper — 564 ms, 4.0%Chirp 3 — 832 ms, 3.2%Chirp 2 — 897 ms, 3.9%STT Rt v6 — 57 ms, 3.9%

Where the errors come from

WER compositionSTT Rt v6's Word Error Rate split by error type · 30-day averageSTT Rt v6's WER split into substitutions, deletions and insertions.
  • STT Rt v63.9%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

STT Rt v6 latency distributionMilliseconds · shared axis across metrics · last 30 daysSTT Rt v6's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to Final Segmentp50 53 ms · Max 132 ms
  • Time to First Tokenp50 1944 ms · Max 5673 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 Segment57 ms35 ms53 ms78 ms91 ms98 ms132 ms1,461
Word Error Rate3.9%0.0%0.0%5.9%12.5%19.0%68.0%1,461
Time to First Token1888 ms1584 ms1944 ms2124 ms2414 ms2724 ms5673 ms1,461

Time to Final Segment by dataset

STT Rt v6 by test conditionMilliseconds · lower is better · best condition firstSTT Rt v6's Time to Final Segment on each benchmark dataset.
  1. WildASR clean54 ms
  2. WildASR clipping56 ms
  3. PipeCat (production)57 ms
  4. WildASR noise gaps58 ms
  5. WildASR far-field58 ms
  6. WildASR phone codec58 ms
  7. WildASR accents62 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
ProductionPipeCat57 ms769
AccentsWildASR62 ms77
CleanWildASR54 ms308
ClippingWildASR56 ms77
Far-fieldWildASR58 ms77
Noise gapsWildASR58 ms76
Phone codecWildASR58 ms77

Strongest condition: WildASR clean at 54 ms · weakest: WildASR accents at 62 ms.

How fast is STT Rt v6?

On Soniox, STT Rt v6 measures mean 57 ms time to final segment (4th of 28) and mean 1888 ms time to first token (20th of 28). Last measured 2026-11-05.

How accurate is STT Rt v6?

On Soniox, STT Rt v6 measures 3.9% word error rate (11th of 30). Last measured 2026-11-05.

Who hosts STT Rt v6?

STT Rt v6 is created by Soniox and served by Soniox. 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