Voice AI BenchmarksBy

CARTESIASPEECH-TO-TEXT114,458 SAMPLES / 30 DAYSLAST RUN SEP 15, 2026, 08:30 UTC

official resource

Ink 2 speech-to-text benchmarks

Ink 2, hosted by Cartesia, measures mean 122 ms time to final segment (10th of 28) and 4.8% 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 .

Ink 2 is Cartesia's real-time speech recognition model.

Time to Final Segment#10 / 28
122ms
Word Error Rate#11 / 30
4.8%
Time to First Token#20 / 26
1827ms

Overview

Cartesia provides Ink through a streaming speech-to-text API with voice activity and an on-premises option.

Ink 2 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
Cartesia
Hosted by
Cartesia
Source
Official API
Licensing
Proprietary
Deployment
On-prem
Region
US
Features
VAD

How Ink 2 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 Ink 2 highlighted.
  1. #1Qwen3 ASR 1.7b39 ms
  2. #2Qwen3 ASR Fast46 ms
  3. #3STT RT v557 ms
  4. #4STT 165 ms
  5. #5Parakeet TDT 0.6B v381 ms
  6. #6Nova 389 ms
  7. #7Nova 292 ms
  8. #8Flux Multilingual98 ms
  9. #9Flux99 ms
  10. #10Ink 2122 ms
  11. #11Whisper Large v3via Baseten125 ms
  12. #12Scribe v2 Realtime133 ms
Show all 28 models
  1. #13Universal 3.5 Pro173 ms
  2. #14Grok STT207 ms
  3. #15Defaultvia Speechmatics209 ms
  4. #16Pulse212 ms
  5. #17Defaultvia Gradium246 ms
  6. #18resonant-1264 ms
  7. #19Whisper Large v3via Together AI295 ms
  8. #20Enhanced299 ms
  9. #21Gemini 3.5 Transcribe Live305 ms
  10. #22Voxtral Mini Transcribe Realtime 2602401 ms
  11. #23GPT Realtime Whisper551 ms
  12. #24GPT-4o mini Transcribe690 ms
  13. #25GPT-4o Transcribe754 ms
  14. #26Chirp 3776 ms
  15. #27Solaria 1805 ms
  16. #28Chirp 2873 ms
median of all models · 208 ms
STT models over the last 30 days, ranked on Time to Final Segment.
#ModelHostTTFSWERTTFTSamples
1Qwen3 ASR 1.7bBaseten39 ms931 ms1,256
2Qwen3 ASR FastNari46 ms1748 ms4,451
3STT RT v5Soniox57 ms1529 ms22,468
4STT 1Inworld AI65 ms1400 ms22,836
5Parakeet TDT 0.6B v3Together AI81 ms1215 ms22,488
6Nova 3Deepgram89 ms1419 ms22,855
7Nova 2Deepgram92 ms1420 ms22,733
8Flux MultilingualDeepgram98 ms1164 ms12,251
9FluxDeepgram99 ms1076 ms12,243
10Ink 2Cartesia122 ms1827 ms22,862
11Whisper Large v3Baseten125 ms912 ms1,252
12Scribe v2 RealtimeElevenLabs133 ms2175 ms22,866
13Universal 3.5 ProAssemblyAI173 ms1041 ms20,253
14Grok STTxAI207 ms—22,853
15DefaultSpeechmatics209 ms1428 ms22,876
16PulseSmallest212 ms2011 ms22,856
17DefaultGradium246 ms1980 ms22,839
18resonant-1Reson8264 ms—22,860
19Whisper Large v3Together AI295 ms1338 ms22,744
20EnhancedSpeechmatics299 ms1492 ms22,876
21Gemini 3.5 Transcribe LiveGemini305 ms1679 ms16,403
22Voxtral Mini Transcribe Realtime 2602Mistral401 ms1851 ms22,583
23GPT Realtime WhisperOpenAI551 ms1814 ms22,812
24GPT-4o mini TranscribeOpenAI690 ms—22,830
25GPT-4o TranscribeOpenAI754 ms—22,831
26Chirp 3Google776 ms5974 ms22,875
27Solaria 1Gladia805 ms1804 ms22,439
28Chirp 2Google873 ms6071 ms22,789
—Nemotron 3.5 ASR StreamingTogether AI—1548 ms22,747
—Universal StreamingAssemblyAI—1514 ms20,235
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 · Ink 2 highlighted · 30-day averagesTime to Final Segment against Word Error Rate for every measured STT model, with Ink 2 highlighted.
Qwen3 ASR 1.7b — 39 ms, 4.1%Qwen3 ASR Fast — 46 ms, 3.2%STT RT v5 — 57 ms, 5.7%STT 1 — 65 ms, 4.4%Parakeet TDT 0.6B v3 — 81 ms, 11.2%Nova 3 — 89 ms, 6.2%Nova 2 — 92 ms, 7.9%Flux Multilingual — 98 ms, 8.8%Flux — 99 ms, 6.7%Whisper Large v3 via Baseten — 125 ms, 5.6%Scribe v2 Realtime — 133 ms, 5.5%Universal 3.5 Pro — 173 ms, 3.1%Grok STT — 207 ms, 4.6%Default via Speechmatics — 209 ms, 5.4%Pulse — 212 ms, 5.3%Default via Gradium — 246 ms, 10.0%resonant-1 — 264 ms, 3.4%Whisper Large v3 via Together AI — 295 ms, 8.5%Enhanced — 299 ms, 4.3%Gemini 3.5 Transcribe Live — 305 ms, 3.9%Voxtral Mini Transcribe Realtime 2602 — 401 ms, 6.2%GPT Realtime Whisper — 551 ms, 5.1%GPT-4o mini Transcribe — 690 ms, 4.9%GPT-4o Transcribe — 754 ms, 4.6%Chirp 3 — 776 ms, 4.1%Solaria 1 — 805 ms, 7.1%Chirp 2 — 873 ms, 4.9%Ink 2 — 122 ms, 4.8%

Where the errors come from

WER compositionInk 2's Word Error Rate split by error type · 30-day averageInk 2's WER split into substitutions, deletions and insertions.
  • Ink 24.8%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Ink 2 latency distributionMilliseconds · shared axis across metrics · last 30 daysInk 2's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to Final Segmentp50 119 ms · p99 186 ms
  • Time to First Tokenp50 1790 ms · p99 3570 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 Segment122 ms104 ms119 ms134 ms152 ms164 ms186 ms22,862
Word Error Rate4.8%0.0%0.0%6.7%14.3%20.0%40.0%22,899
Time to First Token1827 ms1583 ms1790 ms2010 ms2387 ms2593 ms3570 ms22,899

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 daysInk 2's daily median Time to Final Segment over the last 30 days.
Ink 2 · Sep 1: 136 msInk 2 · Sep 2: 133 msInk 2 · Sep 3: 134 msInk 2 · Sep 4: 129 msInk 2 · Sep 5: 121 msInk 2 · Sep 6: 126 msInk 2 · Sep 7: 128 msInk 2 · Sep 8: 126 msInk 2 · Sep 9: 130 msInk 2 · Sep 10: 131 msInk 2 · Sep 11: 128 msInk 2 · Sep 12: 131 msInk 2 · Sep 13: 124 msInk 2 · Sep 14: 134 msInk 2 · Sep 15: 123 ms
Word Error Rate — daily averageDaily average · UTC daysInk 2's daily Word Error Rate over the last 30 days.
Ink 2 · Sep 1: 7.7%Ink 2 · Sep 2: 5.4%Ink 2 · Sep 3: 5.7%Ink 2 · Sep 4: 5.4%Ink 2 · Sep 5: 5.0%Ink 2 · Sep 6: 5.0%Ink 2 · Sep 7: 5.7%Ink 2 · Sep 8: 5.2%Ink 2 · Sep 9: 5.6%Ink 2 · Sep 10: 5.4%Ink 2 · Sep 11: 5.1%Ink 2 · Sep 12: 5.2%Ink 2 · Sep 13: 5.4%Ink 2 · Sep 14: 5.7%Ink 2 · Sep 15: 5.3%

Time to Final Segment by dataset

Ink 2 by test conditionMilliseconds · lower is better · best condition firstInk 2's Time to Final Segment on each benchmark dataset.
  1. WildASR noise gaps120 ms
  2. WildASR clean121 ms
  3. PipeCat (production)122 ms
  4. WildASR clipping123 ms
  5. WildASR accents123 ms
  6. WildASR far-field124 ms
  7. WildASR reverb125 ms
  8. WildASR phone codec125 ms
Time to Final Segment per dataset, with the number of samples behind each figure.
DatasetTTFSSamples
LibriSpeech96 ms1
ProductionPipeCat122 ms11,515
AccentsWildASR123 ms1,116
CleanWildASR121 ms4,578
ClippingWildASR123 ms1,140
Far-fieldWildASR124 ms1,143
Noise gapsWildASR120 ms1,142
Phone codecWildASR125 ms1,138
ReverbWildASR125 ms1,089

Strongest condition: LibriSpeech at 96 ms · weakest: WildASR phone codec at 125 ms.

How fast is Ink 2?

On Cartesia, Ink 2 measures mean 122 ms time to final segment (10th of 28) and mean 1827 ms time to first token (20th of 26). Last measured 2026-09-15.

How accurate is Ink 2?

On Cartesia, Ink 2 measures 4.8% word error rate (11th of 30). Last measured 2026-09-15.

Who hosts Ink 2?

Ink 2 is created by Cartesia and served by Cartesia. Coval measures each hosted endpoint separately.

Limits of this comparison

Ink 2 is measured separately from Cartesia's Sonic synthesis models and ranked only against other STT models.

  • The benchmark does not combine Ink with Sonic, so end-to-end Cartesia agent latency must be evaluated separately.

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