Voice AI BenchmarksBy

SMALLEST.AITEXT-TO-SPEECH45,584 SAMPLES / 30 DAYSLAST RUN SEP 15, 2026, 07:00 UTC

official resource

Lightning v3.1 Pro text-to-speech benchmarks

Lightning v3.1 Pro, hosted by Smallest.ai (Smallest, smallest.ai), measures mean 362 ms time to first audio (18th of 28) and 4.4% word error rate (5th of 28) among TTS systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

Lightning v3.1 Pro is Smallest.ai's streaming text-to-speech model.

Time to First Audio#18 / 28
362ms
TTFA Network Roundtrip#17 / 28
235ms
TTFA Leading Silence#21 / 28
127ms
Word Error Rate#5 / 28
4.4%

Overview

Lightning is built for real-time multilingual synthesis and voice cloning.

Lightning v3.1 Pro 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
Smallest.ai
Hosted by
Smallest.ai
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Multilingual, Voice cloning

How Lightning v3.1 Pro 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 Lightning v3.1 Pro highlighted.
  1. #1vui66 ms
  2. #2Qwen3 TTS Fast71 ms
  3. #3TTS Flash 291 ms
  4. #4Qwen3 TTS 1.7b106 ms
  5. #5Palabra TTS v1115 ms
  6. #6TTS 2181 ms
  7. #7Flash v2.5194 ms
  8. #8Default235 ms
  9. #9TTS Rt v2245 ms
  10. #10TTS RT v1245 ms
  11. #11Mist v3258 ms
  12. #12Sonic 3.5276 ms
  13. #13Phantom Z 3.4 conversational286 ms
  14. #14Aura 2308 ms
  15. #15Coda312 ms
  16. #16S2.1 Pro335 ms
  17. #17Eleven v3 Conversational348 ms
  18. #18Lightning v3.1 Pro362 ms
Show all 28 models
  1. #19S1379 ms
  2. #20Grok TTS397 ms
  3. #21Sonic 3.6423 ms
  4. #22Simba 3.2452 ms
  5. #23Simba 3.0471 ms
  6. #24Chirp 3 HD535 ms
  7. #25Falcon 2545 ms
  8. #26Qwen3 TTS Flash Realtime754 ms
  9. #27S2.1 Pro Free965 ms
  10. #28GPT-4o mini TTS1013 ms
median of all models · 310 ms
TTS models over the last 30 days, ranked on Time to First Audio.
#ModelHostTTFAWERSamples
1vuiFluxions66 ms9,450
2Qwen3 TTS FastNari71 ms2,190
3TTS Flash 2Inworld AI91 ms11,401
4Qwen3 TTS 1.7bBaseten106 ms420
5Palabra TTS v1Palabra115 ms11,385
6TTS 2Inworld AI181 ms11,401
7Flash v2.5ElevenLabs194 ms9,118
8DefaultGradium235 ms9,127
9TTS Rt v2Soniox245 ms11,232
10TTS RT v1Soniox245 ms11,234
11Mist v3Rime258 ms11,387
12Sonic 3.5Cartesia276 ms11,400
13Phantom Z 3.4 conversationalDeepdub286 ms11,396
14Aura 2Deepgram308 ms11,386
15CodaRime312 ms11,392
16S2.1 ProFish Audio335 ms11,400
17Eleven v3 ConversationalElevenLabs348 ms11,389
18Lightning v3.1 ProSmallest362 ms11,401
19S1Fish Audio379 ms11,393
20Grok TTSxAI397 ms11,386
21Sonic 3.6Cartesia423 ms8,210
22Simba 3.2Speechify452 ms11,393
23Simba 3.0Speechify471 ms11,396
24Chirp 3 HDGoogle535 ms11,400
25Falcon 2Murf545 ms11,397
26Qwen3 TTS Flash RealtimeAlibaba754 ms11,226
27S2.1 Pro FreeFish Audio965 ms11,366
28GPT-4o mini TTSOpenAI1013 ms11,370
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: 5th of 28 on Word Error Rate.

Latency vs accuracy

TTFA vs WEREach point is one measured model · Lightning v3.1 Pro highlighted · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model, with Lightning v3.1 Pro highlighted.
vui — 66 ms, 5.5%Qwen3 TTS Fast — 71 ms, 4.2%TTS Flash 2 — 91 ms, 5.4%Qwen3 TTS 1.7b — 106 ms, 5.5%Palabra TTS v1 — 115 ms, 5.8%TTS 2 — 181 ms, 4.6%Flash v2.5 — 194 ms, 6.7%Default — 235 ms, 5.3%TTS Rt v2 — 245 ms, 4.2%TTS RT v1 — 245 ms, 3.9%Mist v3 — 258 ms, 5.6%Sonic 3.5 — 276 ms, 5.9%Phantom Z 3.4 conversational — 286 ms, 6.0%Aura 2 — 308 ms, 5.2%Coda — 312 ms, 5.0%S2.1 Pro — 335 ms, 4.8%Eleven v3 Conversational — 348 ms, 4.4%S1 — 379 ms, 4.9%Grok TTS — 397 ms, 4.9%Sonic 3.6 — 423 ms, 5.3%Simba 3.2 — 452 ms, 4.4%Simba 3.0 — 471 ms, 5.2%Chirp 3 HD — 535 ms, 5.2%Falcon 2 — 545 ms, 5.2%Qwen3 TTS Flash Realtime — 754 ms, 8.9%S2.1 Pro Free — 965 ms, 4.7%GPT-4o mini TTS — 1013 ms, 4.8%Lightning v3.1 Pro — 362 ms, 4.4%

Where the errors come from

WER compositionLightning v3.1 Pro's Word Error Rate split by error type · 30-day averageLightning v3.1 Pro's WER split into substitutions, deletions and insertions.
  • Lightning v3.1 Pro4.4%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Lightning v3.1 Pro latency distributionMilliseconds · shared axis across metrics · last 30 daysLightning v3.1 Pro's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to First Audiop50 336 ms · p99 789 ms
  • TTFA Network Roundtripp50 226 ms · p99 418 ms
  • TTFA Leading Silencep50 106 ms · p99 539 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 Audio362 ms293 ms336 ms383 ms512 ms611 ms789 ms11,401
TTFA Network Roundtrip235 ms215 ms226 ms240 ms260 ms288 ms418 ms11,401
TTFA Leading Silence127 ms64 ms106 ms143 ms258 ms366 ms539 ms11,401
Word Error Rate4.4%0.0%0.0%5.9%18.8%20.0%35.7%11,381

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 daysLightning v3.1 Pro's daily median Time to First Audio over the last 30 days.
Lightning v3.1 Pro · Sep 1: 347 msLightning v3.1 Pro · Sep 2: 341 msLightning v3.1 Pro · Sep 3: 342 msLightning v3.1 Pro · Sep 4: 326 msLightning v3.1 Pro · Sep 5: 339 msLightning v3.1 Pro · Sep 6: 315 msLightning v3.1 Pro · Sep 7: 332 msLightning v3.1 Pro · Sep 8: 341 msLightning v3.1 Pro · Sep 9: 327 msLightning v3.1 Pro · Sep 10: 336 msLightning v3.1 Pro · Sep 11: 342 msLightning v3.1 Pro · Sep 12: 342 msLightning v3.1 Pro · Sep 13: 332 msLightning v3.1 Pro · Sep 14: 339 msLightning v3.1 Pro · Sep 15: 322 ms
Word Error Rate — daily averageDaily average · UTC daysLightning v3.1 Pro's daily Word Error Rate over the last 30 days.
Lightning v3.1 Pro · Sep 1: 3.2%Lightning v3.1 Pro · Sep 2: 3.8%Lightning v3.1 Pro · Sep 3: 4.2%Lightning v3.1 Pro · Sep 4: 4.1%Lightning v3.1 Pro · Sep 5: 5.0%Lightning v3.1 Pro · Sep 6: 5.5%Lightning v3.1 Pro · Sep 7: 3.8%Lightning v3.1 Pro · Sep 8: 4.7%Lightning v3.1 Pro · Sep 9: 4.8%Lightning v3.1 Pro · Sep 10: 4.6%Lightning v3.1 Pro · Sep 11: 4.6%Lightning v3.1 Pro · Sep 12: 4.3%Lightning v3.1 Pro · Sep 13: 4.4%Lightning v3.1 Pro · Sep 14: 4.6%Lightning v3.1 Pro · Sep 15: 3.1%

Time to First Audio by dataset

Lightning v3.1 Pro by test conditionMilliseconds · lower is better · best condition firstLightning v3.1 Pro's Time to First Audio on each benchmark dataset.
  1. Text prompts362 ms
Time to First Audio per dataset, with the number of samples behind each figure.
DatasetTTFASamples
Text prompts362 ms11,401

Strongest condition: Text prompts at 362 ms · weakest: Text prompts at 362 ms.

How fast is Lightning v3.1 Pro?

On Smallest.ai, Lightning v3.1 Pro measures mean 362 ms time to first audio (18th of 28). Last measured 2026-09-15.

How accurate is Lightning v3.1 Pro?

On Smallest.ai, Lightning v3.1 Pro measures 4.4% word error rate (5th of 28). Last measured 2026-09-15.

Who hosts Lightning v3.1 Pro?

Lightning v3.1 Pro is created by Smallest.ai and served by Smallest.ai. Coval measures each hosted endpoint separately.

Limits of this comparison

Lightning is measured separately from the company's Pulse speech-to-text model.

  • Coval does not score cloning fidelity, naturalness or the latency of a combined Smallest.ai agent. Pulse and Lightning measurements cannot simply be added because a production pipeline also adds endpointing, orchestration and network overhead.

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