Even v4: 70 to over 90 languages, voice cloning over 10 seconds of audio, and 150 ms before the first sound in Turbo version

On 28 September, ElevenLabs released its voice synthesis models v4 and v4 Turbo, with a tag-based expression control and a voice that changes language while keeping its identity.

ElevenLabs released two new text-to-speech models on September 28, 2026, Eleven v4 and Eleven v4 Turbo. The linguistic scope increases from 70 languages in the previous generation to more than 90, with the company placing the clearest quality gains on Japanese, Brazilian Portuguese, Mandarin and Cantonese. The model is touted as better on rhythm, emotion and phrasing, with expression tags that can be inserted into the text and stacked to act on tone and flow.

Two capabilities stand out. Instant cloning now requires only about ten seconds of recording time to capture a voice. And a voice recorded in one language can now be expressed in another by adopting the accent of a native speaker while retaining the identity of the original timbre — a dissociation between vocal identity and accent that is of direct interest to dubbing.

The Turbo variant is aimed at chatbots, where latency matters more than thinness: ElevenLabs announces, on its own measurements, a median delay of about 150 milliseconds before the first sound is emitted. Both models are accessible via the company's products and its API.