Eleven v4 and Eleven v4 Turbo
By Steven Van ·
The two models cover 90+ languages, and cloned voices now speak other languages without carrying over their original accent.
ElevenLabs has released Eleven v4 and Eleven v4 Turbo, a new generation of its text to speech model. Eleven v4 covers more than 90 languages and is described as a net upgrade over Eleven v3, with better output quality, emotion, delivery and audio tag handling.
- Eleven v4 (eleven_v4) is the highest-quality variant, meant for content creation, audiobooks and character voiceovers. It is used through the Text to Dialogue API.
- Eleven v4 Turbo (eleven_v4_turbo) is built for real-time use such as conversational agents, with median inference latency of about 100 ms. It is used through the Text to Dialogue WebSocket.
Voice cloning is more accurate. Instant Voice Clones capture a source voice more faithfully from a short sample, and Professional Voice Clones are fully supported. Because v4 reproduces accent, loudness and recording flaws more closely, source audio quality matters more, and ElevenLabs suggests sticking to a single speaking style in training audio. It also says v4 may sound quite different from v3, so comparing both on your own content is worthwhile.
One behaviour change affects cross-language work. When the target language differs from the reference voice's language, v4 generates natural speech in the target language instead of carrying over the original accent. A Korean speaker's clone generating English, for example, gives fluent English without a Korean accent. ElevenLabs calls this deliberate, still being fine-tuned, and says it is exploring a toggle but has no timeline. If a workflow relies on a voice keeping its native accent, test it before migrating.
Both variants use Stability and Similarity settings. The Style and Speed sliders are not available, and SSML is not supported. Audio tags such as [whispering] or [laughing] direct delivery, though ElevenLabs says they do not yet follow instructions perfectly.
