Voice AI BenchmarksBy

CARTESIATEXT-TO-SPEECH8,558 SAMPLES / 30 DAYSLAST RUN OCT 16, 2026, 23:30 UTC

official resource

Sonic 3.6 text-to-speech benchmarks

Sonic 3.6, hosted by Cartesia, measures mean 297 ms time to first audio (21st of 32) and 1.5% word error rate (1st of 32) among TTS systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

Sonic 3.6 is Cartesia's streaming text-to-speech model, released as a stable snapshot on August 27, 2026.

Time to First Audio#21 / 32
297ms
TTFA Network Roundtrip#17 / 32
161ms
TTFA Leading Silence#27 / 32
136ms
Word Error Rate#1 / 32
1.5%

Overview

Sonic 3.6 speaks 44 languages, adding Odia and Urdu, and is fully backwards compatible with Sonic 3.5, including voice clones created on older models.

Cartesia positions the release on naturalness: native-speaker pronunciation, correct confirmation codes and heteronyms, and in-transcript disfluencies.

Sonic 3.6 is tested every day on a fixed set of text prompts, measuring how quickly audible speech starts and how intelligible the result is.

Technical specifications

Made by
Cartesia
Hosted by
Cartesia
Source
Official API
Licensing
Proprietary
Deployment
On-prem
Region
US
Features
Emotion control, Multilingual, Voice cloning

How Sonic 3.6 ranks

Full TTS dashboard
Time to First Audio — every measured TTS modelMilliseconds · lower is better · 30-day averageEvery TTS model ranked on Time to First Audio, with Sonic 3.6 highlighted.
  1. #1vui57 ms
  2. #2Default63 ms
  3. #3Qwen3 TTS Fast67 ms
  4. #4TTS Flash 288 ms
  5. #5Qwen3 TTS 1.7b107 ms
  6. #6Simba 3.0110 ms
  7. #7Palabra TTS121 ms
  8. #8Simba 3.2125 ms
  9. #9drama3-preview184 ms
  10. #10TTS 2186 ms
  11. #11Flash v2.5193 ms
  12. #12Eleven v4 Turbo199 ms
  13. #13Flux TTS204 ms
  14. #14TTS RT v1255 ms
  15. #15TTS Rt v2257 ms
  16. #16Phantom Z 3.4 conversational267 ms
  17. #17Falcon 2281 ms
  18. #18Mist v3287 ms
  19. #19Sonic 3.5292 ms
  20. #20Aura 2297 ms
  21. #21Sonic 3.6297 ms
Show all 32 models
  1. #22Coda332 ms
  2. #23S2.1 Pro343 ms
  3. #24Grok TTS361 ms
  4. #25S1391 ms
  5. #26Lightning v3.1 Pro392 ms
  6. #27Daytona TTS485 ms
  7. #28Aura 2 En565 ms
  8. #29Chirp 3 HD599 ms
  9. #30Qwen3 TTS Flash Realtime719 ms
  10. #31S2.1 Pro Free848 ms
  11. #32GPT-4o mini TTS2568 ms
median of all models · 274 ms
TTS models over the last 30 days, ranked on Time to First Audio.
#ModelHostTTFAWERSamples
1vuiFluxions57 ms1,968
2Gradium TTS Beta 202609Gradium63 ms2,140
3Qwen3 TTS FastNari Labs67 ms2,140
4TTS Flash 2Inworld AI88 ms1,944
5Qwen3 TTS 1.7bBaseten107 ms1,440
6Simba 3.0SpeechifyAI110 ms2,137
7Palabra TTS v1Palabra121 ms1,763
8Simba 3.2SpeechifyAI125 ms2,134
9Drama 3 Preview Us EastFish Audio184 ms386
10TTS 2Inworld AI186 ms1,944
11Flash v2.5ElevenLabs193 ms2,130
12Eleven v4 TurboElevenLabs199 ms1,450
13Flux TTSDeepgram204 ms1,995
14TTS RT v1Soniox255 ms1,808
15TTS Rt v2Soniox257 ms1,805
16Phantom Z 3.4 conversationalDeepdub267 ms1,875
17Falcon 2Murf281 ms2,139
18Mist v3Rime287 ms602
19Sonic 3.5Cartesia292 ms2,140
20Aura 2Deepgram297 ms2,072
21Sonic 3.6Cartesia297 ms2,140
22CodaRime332 ms602
23S2.1 ProFish Audio343 ms2,140
24Grok TTSSpaceXAI361 ms2,135
25S1Fish Audio391 ms2,139
26Lightning v3.1 ProSmallest392 ms2,140
27Daytona TTSGuava485 ms1,344
28Aura 2 EnCloudflare565 ms2,139
29Chirp 3 HDGoogle599 ms2,140
30Qwen3 TTS Flash RealtimeAlibaba719 ms2,139
31S2.1 Pro FreeFish Audio848 ms2,133
32GPT-4o mini TTSOpenAI2568 ms2,018
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: 1st of 32 on Word Error Rate.

Latency vs accuracy

TTFA vs WEREach point is one measured model · Sonic 3.6 highlighted · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model, with Sonic 3.6 highlighted.
vui — 57 ms, 2.1%Default — 63 ms, 1.9%Qwen3 TTS Fast — 67 ms, 1.8%TTS Flash 2 — 88 ms, 1.7%Qwen3 TTS 1.7b — 107 ms, 1.8%Simba 3.0 — 110 ms, 1.6%Palabra TTS — 121 ms, 2.7%Simba 3.2 — 125 ms, 2.0%drama3-preview — 184 ms, 1.9%TTS 2 — 186 ms, 1.6%Flash v2.5 — 193 ms, 1.9%Eleven v4 Turbo — 199 ms, 1.7%Flux TTS — 204 ms, 1.7%TTS RT v1 — 255 ms, 2.5%TTS Rt v2 — 257 ms, 2.4%Phantom Z 3.4 conversational — 267 ms, 1.9%Falcon 2 — 281 ms, 1.7%Mist v3 — 287 ms, 2.1%Sonic 3.5 — 292 ms, 1.6%Aura 2 — 297 ms, 1.9%Coda — 332 ms, 9.5%S2.1 Pro — 343 ms, 1.6%Grok TTS — 361 ms, 1.6%S1 — 391 ms, 1.8%Lightning v3.1 Pro — 392 ms, 1.6%Daytona TTS — 485 ms, 1.7%Aura 2 En — 565 ms, 2.1%Chirp 3 HD — 599 ms, 1.6%Qwen3 TTS Flash Realtime — 719 ms, 3.0%S2.1 Pro Free — 848 ms, 1.5%GPT-4o mini TTS — 2568 ms, 1.8%Sonic 3.6 — 297 ms, 1.5%

Where the errors come from

WER compositionSonic 3.6's Word Error Rate split by error type · 30-day averageSonic 3.6's WER split into substitutions, deletions and insertions.
  • Sonic 3.61.5%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Sonic 3.6 latency distributionMilliseconds · shared axis across metrics · last 30 daysSonic 3.6's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to First Audiop50 290 ms · Max 3066 ms
  • TTFA Network Roundtripp50 155 ms · Max 2932 ms
  • TTFA Leading Silencep50 139 ms · Max 260 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 First Audio297 ms264 ms290 ms323 ms356 ms381 ms3066 ms2,140
TTFA Network Roundtrip161 ms132 ms155 ms180 ms207 ms229 ms2932 ms2,140
TTFA Leading Silence136 ms125 ms139 ms148 ms161 ms182 ms260 ms2,140
Word Error Rate1.5%0.0%0.0%1.4%6.6%12.2%43.8%2,138

Last 30 days

Daily medians from the same measurement runs · gaps are days without qualifying runs.

Time to First Audio — daily p50Line p50 · band p25–p75 · UTC daysSonic 3.6's daily median Time to First Audio over the last 30 days.
Sonic 3.6 · Sep 23: 266 msSonic 3.6 · Sep 24: 320 msSonic 3.6 · Sep 25: 275 msSonic 3.6 · Sep 26: 318 msSonic 3.6 · Sep 27: 278 msSonic 3.6 · Sep 28: 357 msSonic 3.6 · Sep 29: 320 msSonic 3.6 · Sep 30: 312 msSonic 3.6 · Oct 1: 300 msSonic 3.6 · Oct 2: 300 msSonic 3.6 · Oct 3: 316 msSonic 3.6 · Oct 4: 313 msSonic 3.6 · Oct 5: 294 msSonic 3.6 · Oct 6: 322 msSonic 3.6 · Oct 7: 297 msSonic 3.6 · Oct 8: 273 msSonic 3.6 · Oct 9: 279 msSonic 3.6 · Oct 10: 290 msSonic 3.6 · Oct 11: 291 msSonic 3.6 · Oct 12: 268 msSonic 3.6 · Oct 13: 294 msSonic 3.6 · Oct 14: 272 msSonic 3.6 · Oct 15: 310 msSonic 3.6 · Oct 16: 262 ms
Word Error Rate — daily averageDaily average · UTC daysSonic 3.6's daily Word Error Rate over the last 30 days.
Sonic 3.6 · Sep 23: 4.7%Sonic 3.6 · Sep 24: 4.3%Sonic 3.6 · Sep 25: 4.0%Sonic 3.6 · Sep 26: 3.3%Sonic 3.6 · Sep 27: 5.6%Sonic 3.6 · Sep 28: 7.0%Sonic 3.6 · Sep 29: 4.4%Sonic 3.6 · Sep 30: 1.7%Sonic 3.6 · Oct 1: 6.4%Sonic 3.6 · Oct 2: 3.1%Sonic 3.6 · Oct 3: 5.9%Sonic 3.6 · Oct 4: 8.8%Sonic 3.6 · Oct 5: 2.6%Sonic 3.6 · Oct 6: 5.8%Sonic 3.6 · Oct 7: 5.8%Sonic 3.6 · Oct 8: 6.4%Sonic 3.6 · Oct 9: 2.0%Sonic 3.6 · Oct 10: 2.2%Sonic 3.6 · Oct 11: 3.3%Sonic 3.6 · Oct 12: 4.4%Sonic 3.6 · Oct 13: 6.0%Sonic 3.6 · Oct 14: 6.6%Sonic 3.6 · Oct 15: 5.5%Sonic 3.6 · Oct 16: 4.5%

How fast is Sonic 3.6?

On Cartesia, Sonic 3.6 measures mean 297 ms time to first audio (21st of 32). Last measured 2026-10-16.

How accurate is Sonic 3.6?

On Cartesia, Sonic 3.6 measures 1.5% word error rate (1st of 32). Last measured 2026-10-16.

Who hosts Sonic 3.6?

Sonic 3.6 is created by Cartesia and served by Cartesia. Coval measures each hosted endpoint separately.

Limits of this comparison

Coval measures the official endpoint on fixed prompts and separates it from Ink 2 recognition.

  • The sonic-3.6 alias tracks Cartesia's latest stable snapshot, so results can shift when Cartesia moves the alias; sonic-3.6-2026-08-27 is the pinned reference.
  • The benchmark does not score expressive quality, clone similarity or a combined Ink-and-Sonic agent.

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