Fast from the first byte.
56ms
Median time to first audio byte
p50
56 ms
p90
102 ms
Simba 3.2 · Internally measured on our production US East streaming path · 15 Sep 2026. First byte, not audible latency or a guarantee.
Explore streamingIntelligence
that speaks.
Expressive text-to-speech and voice cloning, through one API.
Ready to listen
Transcript Simba 3.2 · Prepared samples
The train slipped out of the city just after sunrise. Maya closed her book and watched the rooftops give way to open fields. For the first time in months, she had nowhere else she needed to be.
Looking for the Speechify reading app? It lives at speechify.com
Speech that keeps upReady for real time.
Ready for real time.
Built for scale.
More speech. Less overhead.
$6/ 1M
Per million characters on Scale
Scale subscription$499 / month
Credit in TTS characters78M / month
One $468 shared monthly credit, equivalent to 78M TTS characters at this rate, not a separate character allowance. Continued usage draws from your top-up balance.
Compare plansExplore the independent benchmarks
Independent blind preference
A frontier voice,
judged without the label.
1,276Elo in the US-accent provider voices view
- 95% confidence interval
- ±18
- Checked
- 20 Sep 2026
- Samples
- 1,350
Artificial Analysis Speech Arena
Elo score and 95% confidence interval
- SpeechifyAISimba 3.2Elo 1276, 95% confidence interval 1258 to 1294.1,276 ±18
US-accent provider voices. The reviewed five-model cohort plus two selected alternatives. Lines show 95% confidence intervals, which can overlap.
Inspect the dated figuresHow to read this benchmark
Artificial Analysis runs blinded pairwise listening tests. Simba 3.2 scored 1,276 Elo with a 95% confidence interval of ±18 in this dated view. Its interval overlaps other models in the cohort, so the result does not establish a guaranteed quality advantage.
The first five rows form the economics comparison cohort. ElevenLabs and Gradium are additional selected reference points. The dated figures are SpeechifyAI's reviewed transcription, not an official AA export. “Provisional” is AA's label, not our judgment.
Production US East
Fast from
the first byte.
56 msmedian time to first byte
- p50
- 56 ms
- p90
- 102 ms
- Measured
- 15 Sep 2026
Streaming response
Time to first audio byte on our production path
Measured first-byte latency, not audible latency, end-to-end agent response time, or a cross-vendor benchmark.
What this latency measures
These figures measure our Simba 3.2 production streaming path in US East from request to the first audio byte. Network distance, input length, client buffering, and the boundary a benchmark chooses can all change the result.
Time to first byte and time to first audible sound are different measurements. We publish this as a dated production observation, not a guarantee or an independent claim that every route is fastest.
Same reviewed cohort
Frontier quality
without frontier cost.
68%lower than the next-lowest modeled price in the reviewed US-voice cohort
- Simba 3.2
- $6.60
- Next lowest
- $20.80
- Unit
- 1M characters
Reviewed five-model cohort, compared on one cost model
USD per 1M characters, zero-based scale
- SpeechifyAISimba 3.2$6.60Artificial Analysis estimates SpeechifyAI Simba 3.2 at $6.60 per million characters.
- InworldRealtime TTS-2 · provisional$20.80Artificial Analysis estimates Inworld Realtime TTS-2 at $20.80 per million characters.
- AlibabaQwen-Audio-3.0-TTS-Plus · provisional$27.60Artificial Analysis estimates Alibaba Qwen-Audio-3.0-TTS-Plus at $27.60 per million characters.
- CartesiaSonic 3.6$49.00Artificial Analysis estimates Cartesia Sonic 3.6 at $49.00 per million characters.
- VUI LabsLuna TTS$80.00Artificial Analysis estimates VUI Labs Luna TTS at $80.00 per million characters.
AA-normalized estimates for the reviewed five-model US-accent provider voices cohort. Subscription assumptions apply. Not a quote or total monthly bill.
Inspect the dated figuresWhy modeled price is not your invoice
Artificial Analysis normalizes API pricing to an estimated per-character measure. For subscriptions, it uses the lowest-cost annual plan with at least 1M characters and assumes 80% usage. This chart compares only the reviewed five-model cohort.
The $6.60 figure is that independent model, not a SpeechifyAI billed rate. Our current subscription fees, included usage, and subsequent usage rates are published separately.
Compare voices. Trust your ears.
The details make the differenceMore than saying it.
More than saying it.
Meaning it.
Identity. Expression. Language.
Explore what makes a voice feel like a voice.
Simba 3.2 · Emotion control
Same words.
Different feeling.
Choose a delivery. Hear how the same voice changes its tone and rhythm.
Neutral.
Beatrice · Delivery sample
The quick brown fox jumped over the lazy dog.
Ready to listen
Simba 3.2 · Voice cloning
A new script.
Your voice.
Start with a consented reference recording. Create a reusable voice for the words that come next.
Explore voice cloning01 / Reference
Original speaker
Ready to listen0:00
02 / Clone
Cloned voice
Ready to listen0:00
Two different recordings. Generated sample: Simba 3.2. Only clone a voice you have permission to use.
Simba 3.0 · Multilingual synthesis
Make yourself
understood.
Hear 6 language samples, each with a voice from its own catalog. These are separate samples, not translations of one recording.
Explore language supportEnglish
en-US
German
de-DE
Spanish
es-MX
French
fr-FR
Italian
it-IT
Portuguese
pt-BR
Prepared Speechify samples. Choose a sample, then press play.Try your own text
The Simba model familyResearch you can hear.
Explore the modelsResearch you can hear.
Models you can build on.
Simba 3.2Expression, in real time.
Our streaming-native English model. Shape emotion, pacing, and delivery without losing the character of the voice.
English · SSML control · Voice cloning
Simba 3.0A voice beyond one language.
Multilingual synthesis for English, German, Spanish, French, Italian, and Portuguese. One streaming-native model.
6 languages · 7 locales · 24kHz audio
From research to your productYour next idea.
Your next idea.
Now with a voice.
Send text. Receive speech. One streaming API, with control over voice, emotion, and timing.
bash
curl https://api.speechify.ai/v1/audio/stream \
-H "Authorization: Bearer $SPEECHIFY_API_KEY" \
-H "Content-Type: application/json" \
-H "Accept: audio/mpeg" \
-d '{
"input": "Your next idea. Now with a voice.",
"model": "simba-3.2",
"voice_id": "beatrice_32"
}' --output speech.mp3Clear from the first characterStart with an idea.
Compare all plansStart with an idea.
Scale when it speaks.
Free
Build and ship on the text-to-speech API. No credit card.
$0/month
Start free- Included every month
- 500K characters
- After that
- Pauses until next month
- Catalog voices, streaming, and SSML
- Commercial use · community support
Starter
For solo devs and early-stage projects.
$10/month
Start Building- Included every month
- 1.9M characters
- After that
- $10 per 1M characters
- Voice cloning, streaming, SSML
- Keep going past the balance with a top-up, no hard cap
Pro
For teams shipping speech in production.
$99/month
Start Pro- Included every month
- 13.5M characters
- After that
- $8 per 1M characters
- Cheaper rate, so the balance goes further
- Voice cloning, no per-plan limit
Scale
For serious production speech traffic.
$499/month
Start Scale- Included every month
- 78M characters
- After that
- $6 per 1M characters
- Our lowest published rate
- Batch synthesis, the only plan that has it
Beyond the standard plans
Enterprise / Custom
For global volume, compliance, and procurement.
No card needed to start. Voice cloning on paid plans. Included usage renews monthly; purchased top-ups never expire.
Which model should I start with?
Use Simba 3.2 for new English speech integrations. For supported languages beyond English, use Simba 3.0. Check each voice's supported models before choosing a pairing.
Explore the model familyCan I use my own voice?
Yes. Voice cloning on paid plans creates a reusable voice from a reference recording. Use a voice you own or have the speaker's permission to clone, and check the voice's supported models before synthesis.
Explore voice cloningCan I try my own text?
The listening demos use prepared recordings. In the SpeechifyAI platform you can try your own text, explore voices, and get an API key for your application.
Explore the developer platformHow does pricing work?
Text-to-speech is metered by character. Starter is $10 per month, including 1.9M characters, with additional usage at $10 per 1M characters from your top-up balance. Free includes 500K characters each month and pauses when that balance runs out.
Compare all plansIs this the Speechify reading app?
SpeechifyAI is our speech technology and developer platform. The app that reads documents, articles, and books aloud is a separate product at speechify.com.
Visit the reading app