ElevenLabs launches Eleven v4 with 10-second voice cloning in more than 90 languages

ElevenLabs released its new text-to-speech model Eleven v4 and a low-latency version, v4 Turbo, on September 28, 2026. The company calls v4 its most emotive model yet and says it is built on an entirely new architecture. Both models are available now in ElevenAgents, ElevenCreative and through the API.

What Eleven v4 does

The model works out tone, pace and emotion from the context of the text. Delivery can also be described in plain language or with inline tags such as [laughs], [whispers], [light rain] or [phone buzzing]. According to the company, v4 follows these tags more accurately than earlier models. In multi-speaker dialogue it reads the whole scene, so speakers react to what was just said. The text a single request can handle rose from 5,000 to 10,000 characters, about 10 minutes of audio. Supported languages went from more than 70 to more than 90; Turkish is on the list in the docs.

Voice cloning from 10 seconds

Instant voice clones can now be made from 10 seconds of audio. Professional voice clones, which v3 did not support, are back. The standout feature is identity across languages: a voice recorded in one language speaks another like a native speaker while keeping the same character. According to the company, the voice no longer drifts back toward its source accent. v4 Turbo is built for voice agents; ElevenLabs says the median time from request to audible speech is around 150 milliseconds. When a caller switches language mid-call, the agent can carry on in the same voice.

For companies producing call center flows, training videos or multilingual content, this makes it easier to publish in many languages with one voice identity. The model runs in the cloud; current prices are on the ElevenLabs site.

Some old habits change with v4: SSML tags are disabled, and natural tags such as [pause] are used for breaks instead. Voice clones made before v4 need to be retrained to work well. ElevenLabs writes that every voice clone requires verified consent from the voice's owner and that generated audio can be detected as AI-made. Being able to clone a voice from a short recording also raises the risk of fake calls and fraud, so companies should not rely on voice alone for phone verification. To weigh the ranking and listener preference claims, it is worth following comparisons from independent researchers.

Sources: ElevenLabs: Introducing Eleven v4, ElevenLabs: Eleven v4 Turbo in ElevenAgents, ElevenLabs: Models documentation.