What is Eleven v4 Turbo?
Eleven v4 Turbo is the real-time text to speech model of the ElevenLabs Eleven v4 family, released on September 28, 2026. It keeps what makes v4 stand out (close voice clones, audio tags, multi-speaker dialogue, 70+ languages) and answers with a median inference latency of about 100 ms. On UnificAlly it runs as elevenlabs/eleven-v4-turbo and is billed per character.
Speed did not cost it quality in listening tests. On the Artificial Analysis Text to Speech arena it holds 1st place with an Elo of 1330, level with the full Eleven v4 at 1328.
What's new in Eleven v4 Turbo
#1 on the Artificial Analysis TTS arena
Eleven v4 Turbo scores an Elo of 1330 in blind listening votes, 1st on the board ahead of Eleven v4 (1328), Qwen-Audio-3.1-TTS-Plus (1300) and Sonic 3.6 (1286), as of October 14, 2026. Eleven v3 Conversational, the previous real-time model, sits at 1209.
About 100 ms median inference
ElevenLabs built Turbo for live use: support agents, assistants and in-game characters that have to answer while the user is still waiting.
Voice clones on the fly
Pass a few audio samples of one speaker as voice and Turbo clones that voice for the request only, then deletes it. Useful for a branded agent voice without a voice library step.
Audio tags and dialogue
Tags like [casual], [annoyed] or [giggle] steer the delivery in place, and a dialogue list renders several speakers into one audio file.
70+ languages, native accents. A voice speaks each target language fluently instead of carrying its original accent into it.
Word timings. One-voice requests can return character alignment and word start and end times for captions or lip-sync.
10,000 characters per request. Plenty for agent replies, and enough for a long scene when needed.
What Eleven v4 Turbo is best for
Voice agents
Support and sales agents that need to answer quickly with a natural voice.
AI assistants
Spoken replies with emotional range, directed by audio tags.
Interactive characters
Game NPCs and companion apps that talk back in real time.
Multilingual support lines
One agent voice speaking 70+ languages without a foreign accent.
Eleven v4 Turbo use cases
Put Turbo behind a support agent so each reply is spoken a moment after the text is ready. Give an in-game character a cloned voice and tag its lines with the mood the scene calls for. Run a multilingual hotline where one branded voice answers in the caller's language. Render short two-speaker exchanges for interactive stories with one dialogue request per beat.
Eleven v4 Turbo limitations
Audio tags are not followed every time yet, and ElevenLabs is still improving tag adherence. There are no style or speed controls and no SSML. Clones copy their source closely, flaws included, so record clean samples. The cross-language accent behavior is always on, so test any voice that must keep its accent in another language. UnificAlly runs Turbo through the same task flow as every model (submit, then poll or receive a webhook), so the ~100 ms figure is model inference time, not the full round trip.
Eleven v4 Turbo vs Eleven v4
The two sit within two Elo points of each other on Artificial Analysis and take the same input. Pick Turbo for anything live or high-volume, where it also costs less per character. Pick Eleven v4 for audiobooks, dubbing and finished voiceovers, which ElevenLabs positions as its highest-quality model.
Eleven v4 Turbo API examples
curl -X POST https://api.unifically.com/v1/tasks \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "elevenlabs/eleven-v4-turbo",
"input": {
"dialogue": [
{ "text": "[casual] Hey, you made it. Did you find the place okay?", "voice": "21m00Tcm4TlvDq8ikWAM" },
{ "text": "[laughing] Only after walking past it twice.", "voice": "AZnzlk1XvdvUeBnXmlld" }
]
}
}'
Poll GET /v1/tasks/{task_id} until status is completed, then read output.audio_url. One-voice requests, voice cloning and every output format are covered at docs.unifically.com.
FAQs
People also ask about Eleven v4 Turbo
Send a POST /v1/tasks request with model set to elevenlabs/eleven-v4-turbo. Put text and a voice ID in input for one voice, or a dialogue list of text and voice pairs for several speakers. You get a task_id back, and the finished task returns an audio_url.
ElevenLabs puts its median inference latency at about 100 ms, not counting network and application time. It is the real-time model of the Eleven v4 family, built for voice agents and interactive characters.
It holds 1st place on the Artificial Analysis Text to Speech arena with an Elo of 1330, just ahead of Eleven v4 at 1328 and Qwen-Audio-3.1-TTS-Plus at 1300 (as of October 14, 2026). The two v4 models share the top spot within their confidence ranges.
ElevenLabs released Eleven v4 Turbo together with Eleven v4 on September 28, 2026.
Yes, the same way Eleven v4 does. Pass audio sample URLs of one speaker as voice to clone a voice for that request only, or send a dialogue list to render several speakers into one audio file.
Each request takes up to 10,000 characters of text. Voice agents usually send a sentence or two at a time, so the limit rarely matters there.
Eleven v4 Turbo is billed per character of text you send, at $0.024/1000 characters. A new account starts with a small free balance, so you can try it before you top up.
Only stability and similarity_boost. Style, speed and speaker boost do not exist on the v4 models, and SSML is not supported, so the delivery comes from the text and its audio tags.
