ElevenLabs introduced Eleven v4 and Eleven v4 Turbo on September 28, 2026. Both turn text into spoken audio, with controls for delivery and support for cloned voices. The distinction is where you use them: v4 targets expressive narration and dialogue; Turbo targets interactive applications where the next spoken reply needs to arrive quickly.
For someone building an audio lesson, that means directing how an explanation sounds. For a developer building a voice tutor, it also means choosing a speech model that fits the pace of a conversation.
Eleven v4 vs. v4 Turbo: which model fits your project?
The model documentation lists separate model IDs for the two variants. Both support more than 90 languages. Their intended workflows provide a more useful starting point than a single ranking for voice quality.| Detail | Eleven v4 | Eleven v4 Turbo |
|---|---|---|
| Model ID | eleven_v4 | eleven_v4_turbo |
| Primary use | Expressive narration and multi-speaker dialogue | Speech for real-time interactions |
| Example project | Audio lesson or character scene | Voice tutor or support agent |
| Languages | 90+ | 90+ |
| Published server-side latency | No figure in the cited latency table | About 100ms median |
With a recorded lesson, generation can finish before the learner presses play. A conversational tutor has to generate speech during the exchange. That difference makes response timing a model-selection factor for the tutor, while a recorded lesson gives you more room to work on delivery and retakes.
Audio tags and IPA: how do you direct Eleven v4 speech?
The release adds natural-language direction through inline audio tags, including[laughs] and [said angrily in French accent]. ElevenLabs also reports improved support for International Phonetic Alphabet (IPA) phonemes, which represent individual speech sounds for custom pronunciation.These controls solve different problems. A tag directs a performance; pronunciation guidance helps a word sound right. In an AI course, for example, a reassuring delivery could help an introductory explanation, while a pronunciation rule could keep a product name consistent across lessons.
Start with a passage from the actual project. Keep the wording and voice the same, then change one delivery instruction. Listening to the two versions makes the effect of that instruction easier to hear.
Voice cloning, speaker consistency and multilingual audio
ElevenLabs says Instant Voice Clones can use as little as 10 seconds of audio, and v4 also supports Professional Voice Clones. The announcement describes improved speaker identity across generated takes and dialogue.For a recurring narrator, identity is what makes the voice recognizable when the mood changes or a sentence is regenerated. An educational series might need the same narrator for an introduction, a worked example and a later correction. Use those samples when choosing the narrator.
Multilingual support also gives a course team a route to localized audio. The language count tells you which markets may be possible; a fluent reviewer can check whether the translated lesson uses the right terminology and sounds natural for its audience.
Eleven v4 Turbo latency: what do 100ms and 150ms measure?
The launch announcement reports approximately 100ms median inference latency and 150ms median time to first speech, with network delay removed from its comparison. The current latency methodology describes the documented 100ms figure as server-side latency: the time from receiving a request to sending the first audio back.That documentation uses WebSocket streaming requests and warm voices, meaning voices already used for a generation. A median is the midpoint. Individual requests can take longer.
A person using your app also waits for the network, any language-model response and audio buffering. For a voice tutor, measure the complete interval from the learner finishing a question to hearing the reply. Use that measurement to locate the delay.
How to access Eleven v4 through ElevenCreative, ElevenAgents and ElevenAPI
ElevenLabs provides three entry points in its platform overview. ElevenCreative is the browser-based production workspace for projects such as voiceovers and dubbing. ElevenAgents is the conversational-agent platform. ElevenAPI provides programmatic access, including Python and TypeScript SDKs.For a first audio lesson, start in the creative workspace with a short script. For a conversational prototype, follow the relevant model documentation or build in ElevenAgents. Use the same sample passage across versions, including a technical term and a change in tone, so the comparison reflects something your project will actually say.




