Skip to content
The Humanness Index™
Built by VapiGitHub

The Humanness Index™

The open benchmark for how human voice AI sounds, so you can pick the model that passes. Built by Vapi.

MethodologyGitHubContactvapi.ai

Code is Apache-2.0. Standings data is CC BY 4.0. Audio clips and source voices are licensed recordings, all rights reserved. Provider logomarks belong to their respective owners and are used nominatively. “The Humanness Index™” name and logo are Vapi trademarks; see TRADEMARKS.md.

  1. Humanness Index™
  2. ElevenLabs
  3. Turbo v2.5

Humanness Index™ · TTS model

ElevenLabs

ElevenLabs Turbo v2.5

by ElevenLabs

Retired from the arena

Turbo v2.5 arrived in July 2024 and extended ElevenLabs' low latency tier from English to 32 languages, making Hindi, French, Spanish, Mandarin and more roughly three times faster than before, with English itself about 25 percent faster.

A real arena clip: a cloned source voice reading a customer support prompt at phone quality.

Turbo v2.5 key stats

Latency (measured)
265 ms1
Languages
322
Price / 1M chars
$503
Released
July 19, 20244
  1. Vapi streaming benchmark (50 trials per model) (checked 2026-06-10) Median of 50 sequential live streaming trials, June 2026; includes network RTT from the benchmark machine.
  2. elevenlabs.io/docs/overview/models (checked 2026-06-10) Low-latency tier default (Turbo/Flash v2.5); Turbo v2 and Flash v2 are English-only, Multilingual v2 lists 29, and Eleven v3 lists 70+, encoded per model.
  3. elevenlabs.io/pricing/api (checked 2026-06-10) ElevenAPI pay-as-you-go: Flash/Turbo $0.05 per 1k characters = $50 per 1M. Eleven v3 and Multilingual v2 bill $0.10 per 1k ($100 per 1M), encoded per model.
  4. elevenlabs.io/blog/introducing-turbo-v25 (checked 2026-06-10)

Background

Turbo v2.5 arrived in July 2024 and extended ElevenLabs' low latency tier from English to 32 languages, making Hindi, French, Spanish, Mandarin and more roughly three times faster than before, with English itself about 25 percent faster. ElevenLabs has since positioned Flash v2.5 as its successor. It was the strongest ElevenLabs model in blind listening tests while it competed on the Humanness Index™, and it is now retired from the arena in favor of one entry per active family.

Sources: elevenlabs.io

At a glance

Turbo v2.5 brought 32 languages to the low latency tier when it launched in July 2024. In our 50 trial streaming benchmark it returned first audio in a median of 265 ms.

Sources: elevenlabs.io

Frequently asked questions

How is Turbo v2.5 tested on the Humanness Index™?
Listeners hear Turbo v2.5 against another model in a blind head to head round, both voices reading the same customer support prompt from the same cloned source voice, and they pick whichever sounds more human. Its Humanness score derives purely from those votes.
How does Turbo v2.5 compare to Flash v2.5?
Flash v2.5 is faster in our measurements (197 ms vs 265 ms median TTFB) and ElevenLabs recommends Flash for new real time builds, but Turbo v2.5 posted the stronger humanness showing in blind tests while it competed.

Keep exploring

ElevenLabsElevenLabsAll ElevenLabs models on the IndexElevenLabsTurbo v2Rank #10 · Humanness 75ElevenLabsFlash v2Rank #9 · Humanness 76ElevenLabsFlash v2.5Rank #14 · Humanness 68ElevenLabsEleven v3Rank #1 · Humanness 96ElevenLabsMultilingual v2Latency 1006 ms

Back to the Humanness Index™

Find the most human-sounding voice for your agent.

Compare the models in blind tests, read the methodology, or get in touch.

Read the methodologyStar on GitHub

Build a TTS model? Add yours to the Index.