ElevenLabs Launches 'Eleven v4' and 'Eleven v4 Turbo' Text-to-Speech AI Models

ElevenLabs has launched Eleven v4, its latest text-to-speech model, alongside Eleven v4 Turbo, a lower-latency version designed for real-time AI agents. The company says the new models are built to make synthetic speech more expressive, consistent and responsive.

Eleven v4 is built to make AI voices sound much more natural

The biggest change with Eleven v4 is how it handles emotion, pacing, tone and context. Instead of simply reading a sentence, the model can interpret how that line should sound and adjust its delivery accordingly.

ElevenLabs says the model can produce dramatic, tender, urgent, comedic or conversational speech while maintaining the identity of the speaker. It also improves conversations between multiple speakers by using the surrounding context instead of treating every line as an isolated generation.

Users can also guide delivery with natural-language instructions and inline audio tags such as [laughs], [said angrily in French accent], [light rain] and [phone buzzing]. The company says Eleven v4 follows these directions more accurately than previous models.

Eleven v4 Turbo targets AI agents that cannot afford to wait

Eleven v4 Turbo is aimed at conversational AI and other applications where latency matters. ElevenLabs says it has a median inference latency of around 100ms, while its median time to first speech is around 150ms.

The company has also expanded multilingual capabilities. Both Eleven v4 and v4 Turbo support more than 90 languages, while a voice recorded in one language can speak another while retaining the original speaker’s identity and adopting a native accent.

Voice cloning gets an upgrade too. ElevenLabs says Instant Voice Clones can now capture a voice with high fidelity from just 10 seconds of audio, while the new models are designed to maintain speaker consistency across longer projects.

Eleven v4 and Eleven v4 Turbo are available now through ElevenAgents, ElevenCreative and the ElevenAPI.