Eleven v3
ElevenLabs' most expressive text-to-speech model with fine-grained emotional control
Eleven v3 is ElevenLabs’ most expressive text-to-speech model, launched in public alpha on June 5, 2025 and moved to general availability in March 2026. It supports over 70 languages and introduces inline audio tags such as [excited], [whispers], [sighs], and [laughing], which let a user script a voice performance rather than just its words, plus multi-speaker dialogue generation for conversations between different voices. The model is built for content where emotional delivery matters, like audiobooks, game characters, and narrative ads, rather than for the low-latency conversational use cases that Flash and Turbo cover. Because it prioritizes expressiveness over speed, ElevenLabs positions it as a separate track from its real-time voice agent models rather than a straight replacement for them.