Eleven v4 Online - Free ElevenLabs v4 TTS Demo
Try Eleven v4 free, no sign-up: type up to 200 characters and hear ElevenLabs' newest TTS model. Audio tags, 90+ languages, real samples, specs and pricing.
Last updated: 2026-10-05
Eleven v4 is ElevenLabs' newest text-to-speech model, released on September 28, 2026. It reads a script the way a voice actor would, follows inline audio tags like [whispers] and [laughs], speaks 90+ languages and takes up to 10,000 characters per request. Type something below and hear it, with no account needed.
Try Eleven v4 now: free, no sign-up
2 free generations a day, up to 200 characters each. Tags in [brackets] are performed, not read aloud.
Eleven v4 at a glance
| Eleven v4 | |
|---|---|
| Made by | ElevenLabs |
| Released | September 28, 2026 |
| Model ID | eleven_v4 (low-latency variant: eleven_v4_turbo) |
| Languages | 90+ (the API lists 85 language codes, 11 more than v3, including Cantonese) |
| Max length per request | 10,000 characters, about 10 minutes of audio |
| Audio tags | Yes, inline and stackable: [whispers], [laughs], [sighs], [excited]… |
| Price on ElevenLabs | Same credit rate as their other TTS models |
| On VoiceSmiths | Pick Eleven v4 in the voice studio; our audio drama studio renders every drama on v4 |
Listen: real Eleven v4 output
Every clip below was generated with eleven_v4 on October 5, 2026, using default settings and one take, with no editing. The transcript under each player is the exact script that was sent, tags included.
English narration:
English with audio tags. The whisper, the nervous question, the laugh and the sigh are all performed, and none of the tags is read aloud:
[whispers] Okay, don't move. I think it's still in the kitchen. [nervous] What if it heard us? [laughs] Relax, it's just the cat. [sighs] You scared me half to death.Mandarin narration:
Mandarin with tags:
[laughs] 你以为我会就这么算了？[whispers] 合同在我手里，三天之内，你会求着来找我。[sighs] 可惜啊，我们本来可以是朋友。Japanese and Brazilian Portuguese, two of the languages ElevenLabs says improved most in v4:
Two characters in one request (text-to-dialogue on v4):
[curious] 你这猴头，又从哪里闯了祸回来？
[laughs] 师父莫急，俺老孙不过是借了他三根毫毛！
[sighs] 借？只怕人家此刻正满山找你呢。
[excited] 找便找，俺一个筋斗，早到了十万八千里外！What's new in Eleven v4
- It reads for meaning. ElevenLabs describes v4 as knowing "who's speaking, what just happened" in a script, so delivery follows the scene rather than each sentence on its own.
- Audio tags go further. Tags can be stacked and sequenced, and v4 follows the order you write them in.
- 90+ languages. The model list we pull from the ElevenLabs API shows 85 language codes for v4 against 74 for v3. The new ones are Cantonese, Burmese, Mongolian, Uzbek, Yoruba, Maltese, Māori, Odia, Tajik, Asturian and Occitan. ElevenLabs singles out Japanese, Brazilian Portuguese, Mandarin and Cantonese as most improved.
- Twice the length per request. 10,000 characters, up from 5,000 on v3, with "context stitching" to keep pacing steady across long scripts.
- The same voice on every retake. Regenerating a line should still sound like the same person, which matters when you fix one sentence in a long chapter.
- A Turbo version for real time.
eleven_v4_turbotargets voice agents; ElevenLabs quotes about 100 ms median inference latency and about 150 ms to first speech.
Audio tags that work on v4
Write the tag in square brackets right before the words it should color. These all worked in our tests:
| Tag | What you hear |
|---|---|
[whispers] | The next words drop to a whisper |
[laughs] | A real laugh, not the word "laughs" |
[sighs] | An audible sigh before the line |
[excited] | Faster, brighter delivery |
[nervous] | Hesitant, unsteady delivery |
[curious] | A questioning, lifted tone |
[sarcastic] | Flat, knowing irony |
[shouts] | Raised voice |
Tags change delivery only. They are never spoken and never show up in subtitles. The full list, with examples, is in our emotion tags cheat sheet.
Eleven v4 vs v3 in short
In our side-by-side test (7 scripts, 3 runs per model, same voices), v4 generated audio 1.6× as fast as v3, with the same accuracy. Neither model read a tag aloud in 42 generations. v4 also doubles the per-request limit and adds 11 languages. The full numbers and every audio pair are on Eleven v4 vs v3: side-by-side test. We also ran the same scripts through Google's new model; see Gemini 3.8 TTS vs Eleven v4.
How to use Eleven v4 on VoiceSmiths
- One voice, any script. Open the voice studio with Eleven v4 already selected, pick a voice, add tags with one click and generate. The credit cost is shown on the button before you spend anything.
- A whole cast. Paste a story into the audio drama studio. It finds the characters, casts a voice for each and directs every line with tags, then renders the chapter on Eleven v4 with subtitles.
- Hear it at length first. All 76 episodes in our audio drama theater, 10.8 hours in total, were rendered on Eleven v4, in Chinese, English and five more languages. Here is one:
Using Eleven v4 through the API
If you call ElevenLabs directly, v4 is one model_id away. Tags go straight into the text:
curl -X POST "https://api.elevenlabs.io/v1/text-to-speech/<voice_id>?output_format=mp3_44100_128" \
-H "xi-api-key: $ELEVENLABS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"text": "[whispers] It works. [laughs] It actually works.", "model_id": "eleven_v4"}' \
--output v4.mp3
One thing we learned while rendering the theater: in a long text-to-dialogue request in Chinese, v4 sometimes dropped the last few lines without an error once a request reached about 1,700 characters. Keeping Chinese, Japanese and Korean requests under 1,200 characters and checking that every line has timestamps fixed it for us.
FAQ
What is Eleven v4? ElevenLabs' newest text-to-speech model. It performs a script the way an actor would, follows inline audio tags such as [whispers] and [laughs], covers 90+ languages and accepts up to 10,000 characters per request.
When was Eleven v4 released? On September 28, 2026, together with Eleven v4 Turbo, a low-latency variant for voice agents.
Is Eleven v4 free? On this page you can generate two clips a day with Eleven v4, up to 200 characters each, without an account. ElevenLabs bills v4 at the same credit rate as its other TTS models, and its own free plan includes 10,000 credits a month. On VoiceSmiths, v4 costs the same as v3; see pricing.
What is the Eleven v4 model ID? eleven_v4 for Eleven v4 and eleven_v4_turbo for Eleven v4 Turbo. Pass it as model_id in a text-to-speech or text-to-dialogue request.
How long can one Eleven v4 generation be? Up to 10,000 characters per request, about ten minutes of audio. Eleven v3 allows 5,000.
Does Eleven v4 support Chinese? Yes. Mandarin is supported, and Cantonese is new in v4. In our test v4 read a Mandarin narration paragraph with no transcription errors in three out of three runs.
Should I use Eleven v4 or Eleven v4 Turbo? Use Eleven v4 for voiceovers, audiobooks, character dialogue and anything you will publish. Use v4 Turbo when latency matters more than nuance, such as live voice agents; ElevenLabs quotes about 100 ms median inference latency for it.
Is VoiceSmiths affiliated with ElevenLabs? No. VoiceSmiths is an independent studio built on the official ElevenLabs API. Eleven v4 and ElevenLabs are trademarks of ElevenLabs.