Humanness Index™ · TTS model
by ElevenLabs
Multilingual v2 is the long standing ElevenLabs quality flagship, launched in August 2023 as the model that took the platform out of beta and extended its voices from English to 29 languages.
Standings as of Jun 13, 2026, 01:14 UTC
A real arena clip: a cloned source voice reading a customer support prompt at phone quality.
Multilingual v2 is the long standing ElevenLabs quality flagship, launched in August 2023 as the model that took the platform out of beta and extended its voices from English to 29 languages. ElevenLabs still describes it as its most lifelike model with rich emotional expression, and it remains the default recommendation for narration, audiobooks, and other pre rendered work where fidelity matters more than speed.
Sources: elevenlabs.io, techcrunch.com
The quality counterpart to the realtime Turbo and Flash families: on the independent Coval benchmark it posts the best word error rate of any ElevenLabs model (3.9 percent) but batch class latency. Our own 50 trial benchmark measured a median of 1006 ms to first audio on its chunked HTTP streaming endpoint, the highest on the Index, which is why it competes here on humanness rather than speed.
Sources: benchmarks.coval.ai, elevenlabs.io
Standings as of Jun 13, 2026, 01:14 UTC
| Rank | Provider | Model | Humanness | Latency |
|---|---|---|---|---|
| #14 | Smallest.ai | Lightning v3.1 | 62 | 420 ms |
| #15 | MiniMax | Speech 2 HD | 62 | 357 ms |
| #16 | ElevenLabs | Multilingual v2 | 59 | 1006 ms |
| #17 | ElevenLabs | Flash v2 | 59 | 226 ms |
| #18 | Cartesia | Sonic 2 | 44 | 159 ms |
The benchmark is open source. Suggest a model, read the methodology, or ask us to put your voice in the arena.
