Eleven v4 ElevenLabs' new voice model: what is new and what the company claims

Expressive delivery and audio tags

You can describe how a line should be delivered in natural language and add inline audio tags such as [laughs], [light rain] or [phone buzzing]. ElevenLabs says v4 follows these tags and direction prompts more accurately than its earlier models.

90+ languages and accents

Both Eleven v4 and Eleven v4 Turbo support more than 90 languages. According to ElevenLabs, a voice recorded in one language can speak others with a native-like accent while keeping its identity, and accent adherence is stronger than before.

Multi-speaker dialogue

ElevenLabs says dialogue between several speakers responds to the context of the scene instead of being assembled line by line. On the official text to speech page, Eleven v4 is listed with multi-speaker dialogue and a 5,000 character limit per request.

Voice cloning

The announcement says cloned voices have better similarity to the source voice, that an Instant Voice Clone can be made from about 10 seconds of audio, and that Professional Voice Clones are now supported.

Pronunciation control

IPA phoneme support, used for custom pronunciations, is described as significantly improved. ElevenLabs also says request stitching for longer content is more reliable in ElevenLabs Studio and the Reader App.

Eleven v4 Turbo

A low-latency variant built for real-time use and optimized to work with ElevenAgents. ElevenLabs reports a median inference latency of about 100 ms and a median time to first speech of about 150 ms in its own tests.

Where you can use it

Eleven v4 is available in ElevenAgents, ElevenCreative and through the ElevenAPI. ElevenLabs invites new users to create a free account to start generating. The announcement does not list pricing or limits, so check the official pricing page.

Third-party ranking

ElevenLabs states that Eleven v4 ranked first on the Artificial Analysis Provider Voice Arena Leaderboard in September 2026. This figure is reported on the ElevenLabs page and was not verified independently here.

Blind listening tests

ElevenLabs reports that about 75% of listeners preferred v4 in blind head-to-head tests against Cartesia Sonic 3.6, Inworld TTS-2 and two Google Gemini TTS models, with win rates between 65% and 81%. These tests were run and described by ElevenLabs.

Latency comparison

The time to first speech comparison (about 150 ms for Turbo, versus 262 to 814 ms for competitors) comes from ElevenLabs' own test, with network latency removed and Turbo measured over WebSocket streaming.