News
Eleven v4 and v4 Turbo: Audio Tags, Voice Clones Return, and a 150 ms Real-Time Model
ยท RenderBob team
ElevenLabs launched Eleven v4 and v4 Turbo on 28 September. v4 is built for produced audio, and Turbo for live calls and agents.
ElevenLabs launched Eleven v4 and v4 Turbo on 28 September. Its changelog for that day lists both. v4 is the quality model for produced audio. Turbo is the real-time model for live calls and agents.
What is reported
Eleven v4 supports more than 90 languages and multiple speakers. It accepts inline audio tags such as [laughs] and [whispers], which ElevenLabs says it follows more reliably than v3. Launch coverage says SSML break tags are not supported, so pauses are controlled through tags, and that a single generation takes up to 10,000 characters. Confirm both limits on the current docs. Professional Voice Clones are available again after being absent from v3. ElevenLabs' v4 Turbo page cites a library of more than 17,500 voices. An older API marketing page cites a smaller library total, so check the number in the product you are actually using. Launch coverage says older Instant and Professional clones need retraining.
Eleven v4 Turbo is the real-time model. ElevenLabs reports median inference latency near 100 ms and median time to first speech of 150 ms, with bidirectional streaming so text can be sent while a language model is still writing. These are company figures. So is the claim of a number-one ranking from Artificial Analysis.
Consent and safeguards
ElevenLabs says every cloned voice needs verified owner consent. Its v4 Turbo page repeats that requirement. Launch coverage also says output is covered by its AI Speech Classifier. That matches the consent points in our voices and transcript-layer article. For a production, get the consent in writing from the voice owner, including the scope of use, and keep the record with the project.
What it means for video work
For narration, dubbing, and character dialogue, audio tags make performance direction more precise than before, such as a whisper on one line. If you use v4 for visual-dubbing workflows, plan time to rebuild existing voices if the retraining requirement holds. Turbo matters less for post-production and more for interactive characters and live agents.
A note on dates
Different outlets give different launch dates, from 28 September to early October. ElevenLabs' own changelog is dated 28 September, which is the date used here.
What to watch
Independent listening tests, how v4 handles long-form narration, and whether the retraining requirement causes problems for existing client voices.
More from the blog
- Higgsfield Says AI Video Now Runs Past 10 Minutes: What an OpenAI Customer Story Does and Doesn't Show
An OpenAI customer story from 13 October says Higgsfield can now produce AI video longer than 10 minutes using GPT-6 Astra. The claims are the company's, and no independent test has been published.
- Disney Bids $20 Million for a License to Character.AI's IP: What the Stalking-Horse Deal Does and Doesn't Say
Character.AI filed for Chapter 11 on 12 October, and Disney Worldwide Services is the opening bidder at $20 million for a non-exclusive, perpetual licence. Higher bids can still come in.