Voice AI BenchmarksBy

SAIDWIRESPEECH-TO-TEXT18,080 SAMPLES / 30 DAYSLAST RUN NOV 10, 2026, 17:30 UTC

Wire-1 speech-to-text benchmarks

Wire-1, hosted by Saidwire, measures mean 35 ms time to final segment (1st of 29) and 3.7% word error rate (11th 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 Wire-1, created by Saidwire and served through Saidwire's own API.

Time to Final Segment#1 / 29
35ms
Word Error Rate#11 / 31
3.7%
Time to First Token#1 / 29
778ms

Overview

Wire-1 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
Saidwire
Hosted by
Saidwire
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US

How Wire-1 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 Wire-1 highlighted.
  1. #1Wire-135 ms
  2. #2Qwen3 ASR 1.7b35 ms
  3. #3STT RT v549 ms
  4. #4Qwen3 ASR Fast52 ms
  5. #5STT Rt v658 ms
  6. #6STT 170 ms
  7. #7Nova 386 ms
  8. #8Nova 289 ms
  9. #9Flux Multilingual101 ms
  10. #10Flux103 ms
  11. #11Universal 3.6 Pro107 ms
  12. #12Universal 3.5 Pro109 ms
Show all 29 models
  1. #13Scribe v2 Realtime117 ms
  2. #14Rumble-2126 ms
  3. #15Whisper Large v3via Baseten128 ms
  4. #16Ink 2142 ms
  5. #17Whisper Large v3via Together AI149 ms
  6. #18Grok STT173 ms
  7. #19Pulse187 ms
  8. #20resonant-1224 ms
  9. #21Linden 1235 ms
  10. #22Default281 ms
  11. #23Daytona STT363 ms
  12. #24Gemini 3.5 Transcribe Live373 ms
  13. #25Solaria 1402 ms
  14. #26Voxtral Mini Transcribe Realtime 2602464 ms
  15. #27GPT Realtime Whisper557 ms
  16. #28Chirp 3839 ms
  17. #29Chirp 2902 ms
median of all models · 128 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Saidwire En 2026 10 30Saidwire35 ms778 ms3,616
2Qwen3 ASR 1.7bBaseten35 ms939 ms13,679
3STT RT v5Soniox49 ms1526 ms27,349
4Qwen3 ASR FastNari Labs52 ms1756 ms27,356
5STT Rt v6Soniox58 ms1883 ms6,349
6STT 1Inworld AI70 ms988 ms27,330
7Nova 3Deepgram86 ms1438 ms24,094
8Nova 2Deepgram89 ms1428 ms24,053
9Flux MultilingualDeepgram101 ms1172 ms24,091
10FluxDeepgram103 ms1094 ms24,099
11Universal 3.6 ProAssemblyAI107 ms1073 ms27,355
12Universal 3.5 ProAssemblyAI109 ms1014 ms27,355
13Scribe v2 RealtimeElevenLabs117 ms2140 ms27,351
14Realtime GlobalReson8126 ms1838 ms27,356
15Whisper Large v3Baseten128 ms918 ms13,222
16Ink 2Cartesia142 ms2030 ms24,037
17Whisper Large v3Together AI149 ms1161 ms7,899
18Grok STTSpaceXAI173 ms—27,332
19PulseSmallest187 ms1967 ms27,350
20resonant-1Reson8224 ms—27,357
21Linden 1Speechmatics235 ms1503 ms27,332
22DefaultGradium281 ms2041 ms27,349
23Daytona STTGuava363 ms5635 ms13,263
24Gemini 3.5 Transcribe LiveGemini373 ms1522 ms27,262
25Solaria 1Gladia402 ms1568 ms27,292
26Voxtral Mini Transcribe Realtime 2602Mistral464 ms1882 ms17,807
27GPT Realtime WhisperOpenAI557 ms1813 ms26,869
28Chirp 3Google839 ms6029 ms27,300
29Chirp 2Google902 ms6091 ms27,354
—Nemotron 3.5 ASR StreamingTogether AI—1568 ms7,825
—Universal StreamingAssemblyAI—1512 ms27,318
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 · Wire-1 highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Wire-1 highlighted.
Qwen3 ASR 1.7b — 35 ms, 2.9%STT RT v5 — 49 ms, 4.7%Qwen3 ASR Fast — 52 ms, 2.6%STT Rt v6 — 58 ms, 3.6%STT 1 — 70 ms, 4.7%Nova 3 — 86 ms, 5.3%Nova 2 — 89 ms, 6.6%Flux Multilingual — 101 ms, 7.5%Flux — 103 ms, 5.7%Universal 3.6 Pro — 107 ms, 2.2%Universal 3.5 Pro — 109 ms, 2.4%Scribe v2 Realtime — 117 ms, 4.2%Rumble-2 — 126 ms, 2.5%Whisper Large v3 via Baseten — 128 ms, 4.4%Ink 2 — 142 ms, 3.4%Whisper Large v3 via Together AI — 149 ms, 7.2%Grok STT — 173 ms, 4.1%Pulse — 187 ms, 4.6%resonant-1 — 224 ms, 2.7%Linden 1 — 235 ms, 3.8%Default — 281 ms, 7.6%Daytona STT — 363 ms, 4.6%Gemini 3.5 Transcribe Live — 373 ms, 2.9%Solaria 1 — 402 ms, 6.2%Voxtral Mini Transcribe Realtime 2602 — 464 ms, 5.4%GPT Realtime Whisper — 557 ms, 4.0%Chirp 3 — 839 ms, 3.2%Chirp 2 — 902 ms, 4.0%Wire-1 — 35 ms, 3.7%

Where the errors come from

WER compositionWire-1's Word Error Rate split by error type · 30-day averageWire-1's WER split into substitutions, deletions and insertions.
  • Wire-13.7%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Wire-1 latency distributionMilliseconds · shared axis across metrics · last 30 daysWire-1's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to Final Segmentp50 34 ms · Max 251 ms
  • Time to First Tokenp50 660 ms · Max 2472 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 Segment35 ms28 ms34 ms39 ms45 ms50 ms251 ms3,616
Word Error Rate3.7%0.0%0.0%6.1%12.5%16.7%100.0%3,616
Time to First Token778 ms286 ms660 ms1261 ms1361 ms1495 ms2472 ms3,616

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 daysWire-1's daily median Time to Final Segment over the last 30 days.
Wire-1 · Nov 6: 34 msWire-1 · Nov 7: 35 msWire-1 · Nov 8: 35 msWire-1 · Nov 9: 33 msWire-1 · Nov 10: 35 ms
Word Error Rate — dailyDaily corpus-level · UTC daysWire-1's daily Word Error Rate over the last 30 days.
Wire-1 · Nov 6: 6.3%Wire-1 · Nov 7: 3.9%Wire-1 · Nov 8: 3.4%Wire-1 · Nov 9: 3.6%Wire-1 · Nov 10: 3.1%

Time to Final Segment by dataset

Wire-1 by test conditionMilliseconds · lower is better · best condition firstWire-1's Time to Final Segment on each benchmark dataset.
  1. WildASR accents26 ms
  2. WildASR noise gaps34 ms
  3. PipeCat (production)34 ms
  4. WildASR clean36 ms
  5. WildASR clipping37 ms
  6. WildASR far-field38 ms
  7. WildASR phone codec38 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
ProductionPipeCat34 ms1,900
AccentsWildASR26 ms191
CleanWildASR36 ms764
ClippingWildASR37 ms189
Far-fieldWildASR38 ms191
Noise gapsWildASR34 ms191
Phone codecWildASR38 ms190

Strongest condition: WildASR accents at 26 ms · weakest: WildASR phone codec at 38 ms.

How fast is Wire-1?

On Saidwire, Wire-1 measures mean 35 ms time to final segment (1st of 29) and mean 778 ms time to first token (1st of 29). Last measured 2026-11-10.

How accurate is Wire-1?

On Saidwire, Wire-1 measures 3.7% word error rate (11th of 31). Last measured 2026-11-10.

Who hosts Wire-1?

Wire-1 is created by Saidwire and served by Saidwire. 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