Listeners rate Cartesia's Sonic 3.6 above every Smallest AI model.
ServiceNow, Decagon, and Quora run their production voice agents on Cartesia. One model, 44 languages, and voice cloning on the public API.
Voice quality
Listeners rate Sonic 3.6 above every Smallest AI model
- Artificial Analysis plays listeners two clips of the same line without naming the model, and rates every model on the board the same way.
- Sonic 3.6 leads Lightning v3.1 Pro, Smallest AI's highest-rated model, by more than 100 points.
Provider Voice Arena Elo
Higher is better
Latency
Sonic 3.6 is faster at every quartile
- Coval calls each provider's public API and times every request to first audio.
- Every quartile of Lightning v3.1 Pro's spread sits above Sonic 3.6's, from the fastest calls to the slowest. This is not a gap that shows up only in the average.
Latency Variation: Distribution of TTFA values
Lower is better
Read the latency measurements
Time to first audio in milliseconds. The box spans the middle 50% of requests; the line marks the median. Whiskers are not absolute minimums or maximums.
- Sonic 3.6: median 307ms; middle 50% 277–352ms; whiskers 206–463ms.
- Lightning v3.1 Pro: median 334ms; middle 50% 301–399ms; whiskers 240–546ms.
Tail latency
Sonic 3.6 answers faster on the slow calls, too
- Coval calls each provider's public API and records how long every request takes to return its first audio.
- At P95, Sonic 3.6 returns first audio in 398ms, against 697ms for Lightning v3.1 Pro.
Latency at P95
Lower is better
Accuracy
Sonic 3.6 makes fewer word errors than Lightning v3.1 Pro
- Coval transcribes every clip each model generates and compares it with the script, so a lower rate means more of the script is spoken as written.
- Sonic 3.6 misreads fewer words than Lightning v3.1 Pro, so names, numbers, and codes come through as written.
Word error rate
Lower is better
Speed
Sonic 3.6 generates faster than Lightning v3.1 Pro
- Artificial Analysis times every model on the board with the same harness, in characters generated per second.
- Sonic 3.6 generates 119.1 characters per second, against 95.5 for Lightning v3.1 Pro.
Characters per second
Higher is better
Reasons why companies choose Cartesia over Smallest AI
Human-like naturalness
Intonation, pacing, pronunciation, emotion, and audio quality that drive higher completion rates
Cartesia1281 EloSmallest AI1171 EloLightning v3.1 Pro1025 EloLightning v3.1October 2026
Latency
Decides whether it feels like a conversation or a phone tree.
Cartesia398ms at P95Smallest AI697ms at P95Lightning v3.1 ProSeptember 2026
Accuracy
Word error rate (WER) on the speech each model generates: whether names, numbers, and terms come through as written.
Cartesia1.85% WERSmallest AI2.22% WERLightning v3.1 ProSeptember 2026
Speed
Headroom over playback, so a long turn never stalls mid-sentence.
Cartesia119.1 characters per secondSmallest AI95.5 characters per secondLightning v3.1 ProSeptember 2026
Languages
Comprehension and trust across a global customer base.
Cartesia44 languages from one model, with native accentsSmallest AI31 languages on Lightning v3.1 ProProfessional voice cloning
A custom voice built from a full recording session.
CartesiaFrom 30 minutes of audio, self-serve on the public APISmallest AIEnterprise plan only
Frequently asked questions
What is Smallest AI?
Smallest AI (smallest.ai) is a voice AI company. It sells Lightning text to speech, Pulse speech to text, and a platform for building voice agents. Its current text-to-speech models are Lightning v3.1 and Lightning v3.1 Pro, which covers 31 languages. On the Artificial Analysis Provider Voice Arena, where listeners compare two clips without knowing which model made either, Lightning v3.1 Pro scores 1171 and Cartesia's Sonic 3.6 scores 1281, as of October 2026.
How is the quality score measured?
It comes from Artificial Analysis, an independent benchmark. Listeners hear two clips of the same line without being told which model produced either, and pick the one they prefer. Sonic 3.6 rates above both Smallest AI models on the board, Lightning v3.1 Pro and Lightning v3.1.
How fast is Cartesia compared with Smallest AI?
At P95, the slowest one call in twenty, Sonic 3.6 returns first audio in 398ms in Coval's independent testing, against 697ms for Lightning v3.1 Pro. Coval calls each provider's public API and times every request.
How accurately does Cartesia read names and numbers?
Sonic 3.6 has a 1.85% word error rate in Coval's testing, against 2.22% for Lightning v3.1 Pro. Coval transcribes each generated clip and compares it with the script.
How many languages does Cartesia support?
44 languages from one model, including English, Spanish, French, German, and Japanese. Accents are calibrated per region rather than bolted onto an English model. Smallest AI lists 31 languages for Lightning v3.1 Pro.
How long does it take to switch from Smallest AI?
One to two weeks for most teams. Cartesia is integrated into the major voice agent platforms and every model is on the public API, so there is no platform to move onto and nothing to unpick later.
More comparisons
Get started today
Talk to an expert.
Connect with a member of our team and learn how Cartesia can help you build world-class voice experiences.
Start building.
Access our models via API and bring a voice agent into production in minutes.