Eleven v4 and Eleven v4 Turbo
ElevenLabs · Voice & Language
Eleven v4 and Eleven v4 Turbo are ElevenLabs text to speech models built for emotional, multi-speaker audio. Turbo keeps that range and cuts latency to about 100 ms for live voice agents.
About Eleven v4 and Eleven v4 Turbo
What Is Eleven v4 and Eleven v4 Turbo
Eleven v4 is the newest ElevenLabs voice model, a fourth-generation text to speech system released on September 28, 2026. It reads a script with attention to who is speaking, what just happened in the scene, and how each line should land. Instead of flattening a paragraph into one steady tone, it shifts pacing, emphasis, and breath to match the writing.
Eleven v4 Turbo is the low-latency side of that expressive speech synthesis system. It keeps the same emotional range, but it's tuned for speed. ElevenLabs puts median inference latency at around 100 ms and time to first speech at about 150 ms. That's fast enough for a phone agent to answer before a caller notices a pause.
Neither model is a general chatbot. You bring the script, choose a voice, and direct the delivery. ElevenLabs rates Eleven v4 first on the Artificial Analysis speech leaderboard and says about 75% of listeners preferred it in blind head-to-head tests, which is a vendor claim worth checking against your own material.
The main limits are practical. A single generation caps at 10,000 characters, billing runs on credits, and any cloned voice needs verified consent from the person it belongs to. Clones made before the v4 launch also need retraining before they perform at full quality. Consent isn't optional.
Getting Started
- Create an ElevenLabs account. The free tier includes 10,000 credits a month, roughly ten minutes of audio.
- Open Text to Speech, then pick Eleven v4 for produced content or Eleven v4 Turbo for a live agent. In the API, switch with the
eleven_v4oreleven_v4_turbomodel ID. - Choose a voice from the 17,500+ library, clone one from a short recording, or generate one from a written description.
- Write the script and add direction in square brackets, such as [whispers], [laughs], or [long pause], wherever the delivery should change.
- Generate the take, then download it as MP3, WAV, PCM, or µ-law depending on where the audio is going.
Product Information
A quick look at Eleven v4 and Eleven v4 Turbo's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- Audiobook and podcast producers
- Game and animation teams
- Product teams building voice agents
Tasks
- Narrating a full manuscript
- Directing a scene with several characters
- Localizing one brand voice across markets
- Staffing a real-time phone line
Scenarios
- Turning a written archive into a listenable series
- Dubbing a course or training video into several languages
- Prototyping a support bot or IVR flow
Key features
Direction written into the script
Eleven v4 reads inline audio tags as performance notes. [whispers], [laughs], [building], [door slams] and similar cues shape tone, pace, and background sound. The model follows tag sequences more reliably than v3, including stacked tags. A scene can be directed in plain language instead of retakes.
Context stitching for long recordings
A generation stops at 10,000 characters, but context stitching carries pacing, tone, and delivery from one chunk into the next, which keeps a long session from sounding like several separate recordings stitched together. An audiobook assembled from dozens of passes sounds like one continuous read. Long books need this. That's what makes long-form work practical rather than a patchwork of takes.
Speaker identity that holds across takes
Regenerating a line used to risk a slightly different person coming back. Eleven v4 keeps speaker identity stable across dialogue, narration, and repeated generations, so fixing one sentence doesn't force you to redo the chapter around it.
Voice cloning from ten seconds of audio
Instant Voice Clones need about ten seconds of clean speech and capture more of the original timbre than earlier models. Ten seconds is enough. Professional Voice Clones, which v3 dropped, are supported again and use the model's full emotional range. Every clone requires verified consent from the voice owner.
More than ninety languages with local accents
Both models cover 90+ languages. A voice recorded in English speaks Japanese, Spanish, or Portuguese with the accent of a native speaker while keeping its identity. Accent adherence lasts through a full generation. That matters for dubbing, and for brands that use one voice across markets.
Turbo for live conversation
Eleven v4 Turbo streams text in and audio out, so it can start speaking before your model finishes the sentence. Median inference latency sits near 100 ms. Does a caller notice the gap? Usually not, and the expressive range still matches the standard model. That combination is what makes a real-time voice agent sound like a person rather than an announcement.
Pros and cons
Pros
- Emotional range that survives a whole scene, not just a single line.
- Audio tags are followed closely enough to replace some manual retakes.
- Professional Voice Clones are back, with the same behavior on both models.
- One cloned voice stays recognizable across more than ninety languages.
- Turbo's latency supports live calls without dropping the expression.
Cons
- 10,000 characters per generation means long books need stitching. Budget extra review time.
- Credit pricing is hard to forecast when you're generating at volume.
- Clones trained before the v4 launch need retraining before they sound right.
- SSML break tags are disabled, so pause control depends on natural-language tags.
Frequently asked questions
Eleven v4 is ElevenLabs' text to speech model for expressive, multi-speaker audio. It interprets tone, pacing, and context in a script, and it accepts inline audio tags that direct how each line is delivered. It launched on September 28, 2026.
Related content
Explore related tools, skills, and articles for Eleven v4 and Eleven v4 Turbo.
Eleven v4 and Eleven v4 Turbo Alternatives
Foley
Hanamori Labs, LLC · Voice & Language · Productivity · AI AssistantFoley is a voice to-do list that turns a spoken sentence into a filed task, complete with project, date and time, then helps you plan the day around what is already there.
HireKey
HireKey · Voice & Language · Productivity · LeaningHireKey is an AI job search copilot built by a former tech recruiter. It checks role fit, tailors resumes, runs interview practice and tracks every application.
Fluence
Yay Software (Yannick Huchard) · Writing · Voice & Language · ProductivityFluence is an Android app that reads PDFs, EPUBs, Word files and web articles aloud, follows each sentence as it speaks, and answers questions about the text on the phone itself.