Request a new model

Tell us which model you want measured. We review every request.

Text-to-Speech

Naturalness

higher is better
field averagetied with bestQwenQwenqwen3-tts-flash1310HumeHumeoctave-21377OpenAIgpt-4o-mini-tts1424RimeRimearcanav31429MiniMaxMiniMaxspeech-2.8-hd1431xAI Grokgrok-tts1495InworldInworldtts-2-flash~1501PalabraPalabratts-v1~1520SmallestSmallestlightning_v3.11544Deepgramflux~1550InworldInworldtts-21561Fish AudioFish Audios2.1-pro~1566Sonioxtts-rt-v11569BlandBlandspeech~1569SpeechifySpeechifysimba-3.21573Cartesiasonic-3.51574Cartesiasonic-3.6~1578Deepgramaura-21584Gradiumdefault~1585ElevenLabseleven_v3_conversational1590GeminiGemini3.1-flash-tts-preview1591Sonioxtts-rt-v2~1606
1300140015001600

Time to first audio

Region
InworldInworldtts-2-flash63ms · 105ms
PalabraPalabratts-v178ms · 94ms
Maya2 Native81ms · 85ms
InworldInworldtts-292ms · 111ms
Cartesiasonic-3.5105ms · 142ms
Cartesiasonic-3.6120ms · 135ms
Sonioxtts-rt-v2134ms · 150ms
Deepgramflux138ms · 160ms
Deepgramaura-2150ms · 163ms
BlandBlandspeech192ms · 262ms
Fish AudioFish Audios2.1-pro206ms · 272ms
Gradiumdefault217ms · 351ms
xAI Grokgrok-tts259ms · 284ms
ElevenLabseleven_v3_conversational268ms · 294ms
RimeRimearcanav3276ms · 302ms
HumeHumeoctave-2424ms · 510ms
QwenQwenqwen3-tts-flash455ms · 474ms
SpeechifySpeechifysimba-3.2474ms · 485ms
OpenAIgpt-4o-mini-tts565ms · 898ms
SmallestSmallestlightning_v3.1654ms · 717ms
GeminiGemini3.1-flash-tts-preview878ms · 1.04s
MiniMaxMiniMaxspeech-2.8-hd1.62s · 1.88s
0ms300ms600ms900ms1.20s

p50 p90

Robustness

higher is better

Whether the voice says the right words for a number, date, currency amount or operator — a different question from drift, which is whether it sounds like itself across repeats.

InworldInworldtts-20.93
InworldInworldtts-2-flash0.93
OpenAIgpt-4o-mini-tts0.93
Sonioxtts-rt-v20.78
SpeechifySpeechifysimba-3.20.70
GeminiGemini3.1-flash-tts-preview0.68
Cartesiasonic-3.50.55
Cartesiasonic-3.60.50
QwenQwenqwen3-tts-flash0.50
RimeRimearcanav30.50
Sonioxtts-rt-v10.49
xAI Grokgrok-tts0.43
Deepgramflux0.41
Deepgramaura-20.41
PalabraPalabratts-v10.30
SmallestSmallestlightning_v3.10.25
ElevenLabseleven_v3_conversational0.24
HumeHumeoctave-20.20
MiniMaxMiniMaxspeech-2.8-hd0.00
Fish AudioFish Audios2.1-pro0.00
Gradiumdefault0.00
BlandBlandspeech0.00
0.000.250.500.751.00

Cost

lower is better

What a vendor charges to synthesize a million characters of text, at its published list rate.

$0$25$50$75$100
~$4
$10
$10
~$13
~$13
$15
$15
$15
$15
~$20
$25
$25
$30
$30
~$33
$40
$45
$50
$50
$58
$100
$100
$100
Maya2 NativeSpeechifySpeechifysimba-3.2QwenQwenqwen3-tts-flashSonioxtts-rt-v1Sonioxtts-rt-v2InworldInworldtts-2-flashFish AudioFish Audios2.1-proxAI Grokgrok-ttsBlandBlandspeechOpenAIgpt-4o-mini-ttsInworldInworldtts-2SmallestSmallestlightning_v3.1Deepgramaura-2PalabraPalabratts-v1GeminiGemini3.1-flash-tts-previewRimeRimearcanav3DeepgramfluxCartesiasonic-3.5Cartesiasonic-3.6GradiumdefaultElevenLabseleven_v3_conversationalMiniMaxMiniMaxspeech-2.8-hdHumeHumeoctave-2

FAQ

Which TTS voice sounds the most natural?
No single voice wins it. In the blind human A/B arena the top 8 systems have overlapping 95% intervals - from Gemini 3.1 Flash TTS at Elo 1591 down to Smallest lightning_v3.1 at 1544, against a field mean of 1500 - so picking a winner inside that band would be reading vote noise.
What is the fastest TTS API for a real-time agent?
Palabra palabra-tts-v1 reached first audio in 72ms p50, the fastest synth measured; Inworld inworld-tts-2 is the fastest fully arena-rated voice at 116ms.
How is TTS naturalness measured on this board?
Each score is an arena Elo from blind human A/B votes on phone-agent lines, fit with Bradley-Terry and published with a bootstrap 95% interval, field mean 1500. Read the interval and not the point: the top 8 voices all overlap, while OpenAI gpt-4o-mini-tts at Elo 1424 sits clear below that band.
Which TTS API is cheapest per million characters?
Maya Maya 2 Native (~$4.0) and Speechify simba-3.2 ($10.0) list the lowest per-character rates on the board. On naturalness: Maya Maya 2 Native carries no naturalness measurement on this board, and Speechify simba-3.2 scored Elo 1573, inside the top arena tie band.
Which TTS voice reads numbers, dates and currency correctly?
Robustness is a 0-1 defect score on numbers, dates, currency and operators. Inworld inworld-tts-2 leads at 0.93.