Back to Gap Map Model components / Speech & audio

ElevenLabs TTS

closed / Overall score: 4.0(strong)

ElevenLabs' hosted text-to-speech models, from Eleven v3 and v3 Conversational (70+ languages, audio tags for emotional control) to the low-latency Flash v2.5, sold through its API and creator apps.

A closed comparator. ElevenLabs' speech-to-text model, Scribe, is a separate surface and is not included here.

Openness

1 high confidence

1.0

weights: closed
data: closed
code: closed
license: Proprietary(API only)

ElevenLabs' voices are served only through its API and apps. No weights, training data or code are published.

https://artificialanalysis.ai/text-to-speech/leaderboard recorded 2026-09-27
ElevenLabs entries carry "openWeights":false.
https://elevenlabs.io/docs/overview/models recorded 2026-09-27
Models overview lists Eleven v3 Conversational, Multilingual v2 and Flash v2.5 as API models with language counts and latency; no weights or self-hosted download are offered.

Adoption

4 low confidence

4.0

ElevenLabs publishes no usage count for its API. Its product page claims more than a million users, a cumulative figure rather than an active one, and millions of words generated every minute; the level reads that claim and is not a measurement.

https://elevenlabs.io/text-to-speech recorded 2026-09-27
Page text: "Trusted by 1M+ users" and "Millions of words generated every minute".

Capability

4 medium confidence

4.0

One of the reference commercial voice services, a little behind the arena's current top ten, which Cartesia and Google lead.

https://artificialanalysis.ai/text-to-speech/leaderboard recorded 2026-09-27
Entries "name":"v3 Conversational" with "elo":1197.04 and "name":"Eleven v3" with "elo":1168.29, creator ElevenLabs.
https://elevenlabs.io/docs/overview/models recorded 2026-09-27
"Eleven v3 Conversational"; "70+ languages supported".

Verified 2026-09-27