Even v4 is the latest voice synthesis model d ElevenLabs launched on September 28, 2026. It comes in two versions: Even v4, thought for quality (audiobooks, advertisements, games, voiceover), and Even v4 Turbo, thought for speed (voice agents, live applications). More than 90 languages are announced, French is listed in the template list, and you are driving the tone with tags written in the text. This is what official sources say, what the model really changes for a Francophone creator, and what it costs.
Released on 25.10.2026 on the official pages of ElevenLabs. This page contains partner links.
Essentials
- Two models:
eleven_v4for quality,eleven_v4_turbofor real time (approximately 100 ms of median inference latency, depending on ElevenLabs). - More than 90 languages, including French (« English (fr) » in the list of document templates).
- Tags in the text for laughter, tone or noise, more natural language instructions and better multi-voice dialogue.
- Cloning more faithful: An instant clone is created from about 10 seconds d
- Limitations to know: no SSML, no Style and Speed sliders, maximum 10,000 characters per query.
- Price: free plan at 10,000 credits per month, then from 6 $ per month (Starter) with commercial license and instant cloning.
Hear Even v4 before deciding: create a free account on ElevenLabs and paste your own text.
Try ElevenLabsWhat is Eleven v4?
ElevenLabs presents Even v4 as its most advanced voice synthesis model and « the first of a new generation ». The official documentation indicates that it improves the quality of rendering, accuracy of voices, emotion, audio tags and language coverage compared to Eleven v3. It is available in the three large bricks of the platform: ElevenLabs for creation, ElevenAgents for voice agents, and API.
The changelog sets technical benchmarks: model identifiers are eleven_v4 and eleven_v4_turbo, and SDK JavaScript and Python in version 2.71.0 added support for the voices of Agent Eleven v4. On October 5, 2026, the default model of agents changed from eleven_flash_v2 to eleven_v4_turbo. On October 19, 2026, real-time voice translation also adopted v4 Turbo by default.
Even v4 or Even v4 Turbo: what difference?
| Criteria | Even v4 | Even v4 Turbo |
|---|---|---|
| Username | eleven_v4 | eleven_v4_turbo |
| Purpose | Audiobooks, ads, characters, voiceover, content creation | Conversational Agents, Interactive Applications |
| Latency | Not specified | Approximately 100 ms of median inference (excluding network) |
| API access | Create speech, Stream speech, Create dialogue, Stream dialogue | Text to WebSocket Dialogue |
| Languages | Over 90 | Over 90 (same list) |
For a video YouTube a podcast or a voiceover Even v4 counts. Turbo focuses on developers who connect a voice to an agent: ElevenLabs mentions a median time before the first speech of about 150 ms, measured in September 2026 in WebSocket, excluded network. These are the numbers of the editor, we did not measure them.
What are the new features of Eleven v4?
According to the official announcement and documentation, the new features are summarized in six points.
- Audio tags in text. Markers such as [laughs], [light groove] or [phone buzzing] are written to trigger a laugh, ambience or noise in the right place.
- Direction in natural language. It describes how the sentence should be said, rather than tinkering with punctuation.
- Dialogue with several voices. The replicas react to the context of the scene, instead of being lines stuck end to end.
- Improved IPA phonetics. Useful for proper names, marks and technical terms that the model pronounces from the wrong.
- Multilingual voices. A voice recorded in one language can speak another language with a native accent, rather than keeping the original accent.
- Longer, more reliable texts. The « request stitching », which chain several requests for long content, is announced as much more stable.
In terms of results, ElevenLabs states that approximately 75% of listeners preferred v4 in blind tests against competitors (the range displayed ranges from 65% to 81%), and quotes a ranking #1 on the Provider Voice Arena of Artificial Analysis in September 2026. These are published results by the publisher, to be taken as such. The only way to know if the voice suits you is to listen to your own text.
Does Even v4 speak French?
Yes. Document Templates Page ElevenLabs list « English (fr) » among the more than 90 languages supported by Even v4 and v4 Turbo. A detail to know: the announcement article does not detail the list of languages, it only speaks of « more than 90 ». It is the documentation that slices for French.
Two rules of the model count in practice. When the language of the text is that of the reference voice, the original accent is retained. When it differs, the model produces a fluid speech in the target language without repeating the reference accent. For a voiceover In French, choose a Francophone voice, and keep foreign voices in cases where you want an accent. To go further on the French voices, our guide of the best AI of the voiceover in French compares options.
What about voice cloning with Eleven v4?
ElevenLabs indicates that an instant vocal clone is created from about 10 seconds d Professional vocal clones are also supported, with a cursor « Clone Strength » The lower the value, the smoother and more flexible the voice, the higher it is, the closer it remains to the original recordings. The documentation recommends starting at 0% for most votes.
Beware of the legal framework: clone only your own voice or that of a person who has given his consent. Our guide of voice cloning with ElevenLabs details the procedure to be followed.
What are the limits of Eleven v4?
The documentation is clear on what the model does not do:
- SSML is not supported (audio tags and text instructions replace it);
- Style and Speed sliders are not available, only Stability and Similarity are available;
- The limit is 10,000 characters per request, i.e. approximately 10 minutes of audio after the template page;
- the cost of credits per character is not detailed on the official pages we have consulted: for Eleven v3, the official help indicates 1 credit per character on the site, and nothing equivalent is published for v4.
Before launching a big project, generate an extract of a few sentences and see how many credits it consumes in your account.
How much does Eleven v4 cost? Rates charged on 25.10.2026
Even v4 does not have a separate rate in the subscription: it is used with the credits of its plan. The formulas on the official tariff page on October 25, 2026 (monthly billing display):
| Plan | Price per month | Appropriations per month | To be retained |
|---|---|---|---|
| Free | 0 $ | 10 000 | No commercial license or cloning |
| Starter | 6 $ 1 month to 1 month $ for promotion until 1 November) | 30 000 | Commercial license, instant cloning |
| Creator | 22 $ 1st month to 11 $) | 121 000 | Professional Cloning Plus |
| Pro | 99 $ | 600 000 | Large volumes |
| Scale | 299 $ | 1.8 million | Teams (3 seats) |
| Business | 990 $ | 6 million | Teams (10 seats) |
Promotions change quickly, check them on the official page before paying. Our page ElevenLabs prices details each formula and the calculation of credits.
Is Even v4 free?
The Free plan ElevenLabs offers 10,000 credits per month, without commercial license or cloning. It remains to be seen whether Even v4 is accessible here: the official tariff page does not clearly confirm this in what we were able to read, and some third-party sites say yes without giving numerical limits. So we cannot say that. The simplest way is to open a free account, see if Eleven v4 appears in the template list, and run a short trial. For details of the free quota, see ElevenLabs free: what you can do with 10,000 credits.
Even v4 or Even v3 : Which Should You Choose ?
Even v3, passed in general availability on February 2, 2026, introduced expressive audio tags. Even v4 repeats this logic with better language coverage, more faithful cloning and natural language direction. For a new project of voiceover or audiobook, v4 is the logical choice, especially as ElevenLabs Puts it on the whole platform. If you already have a workflow that works well in v3 or Multilingual v2, there is no need to migrate: just listen to a control passage in each model before switching.
Who is Eleven v4 interesting for?
- Video and podcast creators: voiceover expressive, laughter and ambiance by beacons, several voices in one scene.
- Authors and Publishers: audiobooks, with long text processing announced as more reliable.
- Developers: v4 Turbo for vocal agents, with low latency.
- Marketing teams: versions of the same advertisement in several languages, with a coherent voice.
If you are looking for other options, see our alternatives to ElevenLabs and comparison ElevenLabs against Murf AI. For a review on the platform, our review ElevenLabs.
How to try Eleven v4 in three steps
- Create an account on ElevenLabs (The free plan is enough to start with).
- Open the voice synthesis, choose the Even v4 model if it is proposed, then a francophone voice.
- Paste a few sentences, add one or two tags like [laughs] or a tone statement, generate, and then compare with the same text in another template.
Verdict: If the voice counts in your content, Eleven v4 deserves a few minutes of listening with your own text.
Test Even v4 on ElevenLabsFAQ on Even v4
When did Eleven v4 come out?
ElevenLabs launched Eleven v4 and Eleven v4 Turbo on 28 September 2026. The announcement was updated on 19 October 2026.
What is the difference between Eleven v4 and Eleven v3?
According to official documentation, v4 improves output quality, voice accuracy, emotion, audio tags and language coverage. The publisher does not publish a precise numerical comparison between the two models.
Does Even v4 support French?
Yes. French is included in the list of more than 90 languages on the document template page ElevenLabs, for Even v4 as for v4 Turbo.
Is Eleven v4 available in the free plan?
This is not confirmed by the official pages we have read. The free plan offers 10,000 credits per month, but without a commercial license. Check your account to see if the template is available.
Can we clone his voice with Even v4?
Yes. Instant cloning takes about 10 seconds of audio and starts with the Starter 6 plan $ per month. Professional cloning is supported from the Creator plan.
Does Even v4 manage SSML?
No. SSML is not supported. The diction is controlled with audio tags, natural language instructions and IPA phonetic notation.
Sources: official announcement of Eleven v4, page « What is Eleven v4? », list of models, changelog and tariff page, recorded on 25.10.2026.
