Voice AI BenchmarksBy

GRADIUMTEXT-TO-SPEECH16,548 SAMPLES / 30 DAYSLAST RUN SEP 4, 2026, 20:30 UTC

Default text-to-speech benchmarks

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

Time to First Audio#5 / 26
236ms
TTFA Network Roundtrip#13 / 26
230ms
TTFA Leading Silence#2 / 26
6ms
Word Error Rate#18 / 26
5.2%

Overview

Default 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
Gradium
Hosted by
Gradium
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Multilingual, Voice cloning

How Default 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 Default highlighted.
  1. #1vui110 ms
  2. #2Palabra TTS v1116 ms
  3. #3TTS Flash 2118 ms
  4. #4TTS 2176 ms
  5. #5Default236 ms
  6. #6Mist v3257 ms
  7. #7TTS Rt v2258 ms
  8. #8TTS RT v1264 ms
  9. #9Sonic 3.5274 ms
  10. #10Phantom Z 3.4 conversational302 ms
  11. #11Coda319 ms
  12. #12Aura 2322 ms
Show all 26 models
  1. #13Flash v2.5357 ms
  2. #14S2.1 Pro363 ms
  3. #15Eleven v3 Conversational396 ms
  4. #16Grok TTS417 ms
  5. #17S1421 ms
  6. #18Simba 3.2447 ms
  7. #19Lightning v3.1 Pro463 ms
  8. #20Sonic 3.6465 ms
  9. #21Simba 3.0469 ms
  10. #22Chirp 3 HD518 ms
  11. #23Falcon 2549 ms
  12. #24Qwen3 TTS Flash Realtime685 ms
  13. #25S2.1 Pro Free902 ms
  14. #26GPT-4o mini TTS1033 ms
median of all models · 360 ms
TTS models over the last 30 days, ranked on Time to First Audio.
#ModelHostTTFAWERSamples
1vuiFluxions110 ms10,720
2Palabra TTS v1Palabra116 ms13,690
3TTS Flash 2Inworld AI118 ms11,379
4TTS 2Inworld AI176 ms13,828
5DefaultGradium236 ms4,137
6Mist v3Rime257 ms13,824
7TTS Rt v2Soniox258 ms11,509
8TTS RT v1Soniox264 ms13,829
9Sonic 3.5Cartesia274 ms13,828
10Phantom Z 3.4 conversationalDeepdub302 ms10,041
11CodaRime319 ms13,819
12Aura 2Deepgram322 ms13,803
13Flash v2.5ElevenLabs357 ms8,559
14S2.1 ProFish Audio363 ms12,882
15Eleven v3 ConversationalElevenLabs396 ms9,410
16Grok TTSxAI417 ms13,823
17S1Fish Audio421 ms12,893
18Simba 3.2Speechify447 ms13,821
19Lightning v3.1 ProSmallest463 ms13,814
20Sonic 3.6Cartesia465 ms3,220
21Simba 3.0Speechify469 ms13,825
22Chirp 3 HDGoogle518 ms13,829
23Falcon 2Murf549 ms12,885
24Qwen3 TTS Flash RealtimeAlibaba685 ms13,791
25S2.1 Pro FreeFish Audio902 ms13,773
26GPT-4o mini TTSOpenAI1033 ms13,816
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

TTFA vs WEREach point is one measured model · Default highlighted · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model, with Default highlighted.
vui — 110 ms, 7.4%Palabra TTS v1 — 116 ms, 5.8%TTS Flash 2 — 118 ms, 5.0%TTS 2 — 176 ms, 4.7%Mist v3 — 257 ms, 6.0%TTS Rt v2 — 258 ms, 3.9%TTS RT v1 — 264 ms, 3.7%Sonic 3.5 — 274 ms, 6.0%Phantom Z 3.4 conversational — 302 ms, 6.3%Coda — 319 ms, 5.1%Aura 2 — 322 ms, 5.2%Flash v2.5 — 357 ms, 6.6%S2.1 Pro — 363 ms, 4.6%Eleven v3 Conversational — 396 ms, 4.3%Grok TTS — 417 ms, 4.7%S1 — 421 ms, 4.9%Simba 3.2 — 447 ms, 4.4%Lightning v3.1 Pro — 463 ms, 4.3%Sonic 3.6 — 465 ms, 5.0%Simba 3.0 — 469 ms, 5.2%Chirp 3 HD — 518 ms, 5.1%Falcon 2 — 549 ms, 5.2%Qwen3 TTS Flash Realtime — 685 ms, 8.8%S2.1 Pro Free — 902 ms, 4.6%GPT-4o mini TTS — 1033 ms, 4.7%Default — 236 ms, 5.2%

Where the errors come from

WER compositionDefault's Word Error Rate split by error type · 30-day averageDefault's WER split into substitutions, deletions and insertions.
  • Default5.2%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Default latency distributionMilliseconds · shared axis across metrics · last 30 daysDefault's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to First Audiop50 214 ms · p99 386 ms
  • TTFA Network Roundtripp50 213 ms · p99 359 ms
  • TTFA Leading Silencep50 0 ms · p99 72 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 First Audio236 ms212 ms214 ms237 ms331 ms349 ms386 ms4,137
TTFA Network Roundtrip230 ms212 ms213 ms217 ms329 ms348 ms359 ms4,137
TTFA Leading Silence6 ms0 ms0 ms0 ms28 ms49 ms72 ms4,137
Word Error Rate5.2%0.0%0.0%7.7%18.8%21.4%32.0%4,137

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 daysDefault's daily median Time to First Audio over the last 30 days.
Default · Aug 25: 214 msDefault · Aug 26: 214 msDefault · Aug 27: 214 msDefault · Aug 28: 213 msDefault · Aug 31: 214 msDefault · Sep 1: 214 msDefault · Sep 2: 215 msDefault · Sep 3: 214 msDefault · Sep 4: 214 ms
Word Error Rate — daily averageDaily average · UTC daysDefault's daily Word Error Rate over the last 30 days.
Default · Aug 25: 4.6%Default · Aug 26: 5.6%Default · Aug 27: 5.2%Default · Aug 28: 6.0%Default · Aug 31: 5.0%Default · Sep 1: 5.3%Default · Sep 2: 4.8%Default · Sep 3: 4.9%Default · Sep 4: 5.0%

Time to First Audio by dataset

Default by test conditionMilliseconds · lower is better · best condition firstDefault's Time to First Audio on each benchmark dataset.
  1. Text prompts236 ms
Time to First Audio per dataset, with the number of samples behind each figure.
DatasetTTFASamples
Text prompts236 ms4,137

Strongest condition: Text prompts at 236 ms · weakest: Text prompts at 236 ms.

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