WTF AGENTS
Guide 26 / 37

WTF is ElevenLabs

The voice you can't tell from real. Now answering the phone.

The voice of the agentic economy. Text-to-speech that crossed the "I can't tell" line, cloning from a short sample, generation in seventy languages — and now the voice agents answering the phone at Klarna and Revolut. What it is, what it costs, and the catches.

The one-liner

ElevenLabs makes AI voices: text-to-speech good enough for audiobooks and film, cloning from a short sample, dubbing that keeps the speaker's voice in another language, and, increasingly, conversational agents that answer a phone line and hold the call.

For a creator it is the voice seat in the stack. For an enterprise it is the reason a support line can answer in thirty languages at 3am. For the agentic economy it is the layer that gives every agent in this series a mouth.

By the numbers — September 2026

$11Bvaluation at the $500M Series D led by Sequoia, February 2026 — up from $3.3B a year earlier
~$22Bvaluation reportedly under discussion for an employee share sale this summer, no new capital raised
$500Mannual recurring revenue passed in April 2026, from $350M at the end of 2025 — $100M added in one quarter
60%+of the Fortune 500 have employees using the platform, per the company — up from 41% in 2024
35MUS customers whose first-line phone support at Klarna is an ElevenLabs voice agent
70+languages for generation on Eleven v3; dubbing runs about thirty
$0 → $299per month: free tier of 10,000 characters through to the Scale plan; Starter $6, Creator $22

What it actually does

Text-to-speech — Type or paste text; get natural speech in a library voice or your own. The v3 models handle emotion, pacing and emphasis from context and from simple tags, which is what moved it past "robot reading."

Voice cloning — Instant cloning from a minute of audio; professional cloning from more, for a voice indistinguishable from the original. Your own voice, or anyone's who has consented in writing.

Dubbing — A video in one language becomes the same video in another, in the original speaker's voice, with timing preserved. Seventy-plus languages in one pass.

Sound effects and music — Describe a sound, get a sound. Background beds for video without a licence hunt.

ElevenReader — The consumer app: any text, article or book read aloud in a chosen voice. Spotify now uses ElevenLabs to let authors produce audiobooks directly.

ElevenAgents — The enterprise product and the growth engine: voice and chat agents that take calls, answer support, qualify sales leads and run hiring screens, with the testing, monitoring and integrations a call centre needs. Revolut runs support for four million customers on it across the UK and Europe; Klarna put it in front of 35 million Americans.

Two years ago ElevenLabs sold a better voice. Now it sells the agent that uses the voice, and that is where the revenue went.

What it costs

PlanPriceWhat you get
Free$010,000 characters a month, attribution required, no cloning
Starter$6/mo30,000 characters, instant cloning, commercial use
Creator$22/mo100,000 characters, professional cloning, higher-quality audio. The tier most people mean by "ElevenLabs"
Pro$99/mo500,000 characters, higher limits, usage-based overage
Scale$299/mo2M characters for teams and studios
APIPer characterRoughly $0.06–0.12 per 1,000 characters depending on model
ElevenAgentsCustomEnterprise voice agents, priced per minute and by contract

The unit is characters, not minutes: a thousand characters is roughly a minute of speech. A weekly ten-minute podcast burns about 10,000 characters an episode; a full audiobook is a million or more. Dubbing and cloning consume faster. Annual billing takes about 17% off.

ElevenLabs vs the field

ElevenLabsOpen-source modelsBig-lab voices
QualityThe benchmarkClose, and closingGood, less expressive
CloningYes, instant and proYes, self-hostedLimited or none
Dubbing~30 languages, one passDIY pipelinePartial
Price$6–299/mo, API per characterFree to run, your hardwareBundled in the plan
AgentsElevenAgents, enterprise-gradeBuild your ownInside ChatGPT, Gemini, Muse
Lock-inVoices live on the platformNoneWith the lab

The open models (Fish Audio, Kokoro and others) are why ElevenLabs cannot raise its API prices much; the enterprise agents are why it does not need to.

The catches

  • Voice rights. Cloning a voice needs documented consent from its owner. Your own is yours; a client's, a colleague's or a celebrity's is not. ElevenLabs verifies professional clones and faces biometric-privacy litigation in the US; the legal exposure for misuse sits with you.
  • Character maths. Long-form audio eats plans fast, and overage on Pro and above is usage-billed. Estimate the month in characters before choosing a tier.
  • Your voices live there. A professional clone is an asset on ElevenLabs' servers. Export the audio you make; the voice itself does not travel.
  • Detection is a moving target. "Can't tell it's AI" cuts both ways. Platforms and regulators are adding disclosure rules for synthetic voice; label it where required.
  • The agent business is not the creator business. Enterprise voice agents are where the company's attention and revenue are going. Creator features keep shipping, but the roadmap is a call centre's.

Who it is for

Yes — Anyone who publishes spoken content: podcasts, video voice-over, courses, audiobooks, multilingual anything. Buy the Creator tier and stop thinking about voice.

Not yet — Anyone whose use needs a voice they do not own the rights to, or who wants to run everything on their own hardware (open models).

No — Anyone who only needs occasional read-aloud (the free tier, or the voice built into their chat app).

Why it matters

Every agent in this series types. ElevenLabs is what lets them speak, and the moment an agent can hold a phone call it stops being software people configure and becomes staff people talk to. That is why the fastest-growing part of an audio company is customer support, and why a voice startup founded in 2022 is worth more than most airlines.

Glossary

ElevenLabs
AI voice company founded 2022 by Mati Staniszewski and Piotr Dąbkowski. Text-to-speech, cloning, dubbing, voice agents. $11B valuation.
Text-to-speech (TTS)
Generating spoken audio from written text.
Voice cloning
Reproducing a specific person's voice from a sample. Instant (short sample) or professional (longer, verified).
Dubbing
Re-voicing spoken content in another language, preserving the original speaker's voice and timing.
ElevenAgents
The enterprise platform for voice and chat agents that take calls and hold conversations.
ElevenReader
The consumer read-aloud app.
Character
The billing unit. Roughly 1,000 characters per minute of speech.
Eleven v3
The current generation of models, with emotion and expressiveness driven by context and tags.
Consent
Written permission from a voice's owner, required before cloning anyone but yourself.

Liked this? Go deeper.

ElevenLabs is the voice. Here is the rest of the creative stack, and the agents that use it.

Get the rest of the series

Every WTF Agents guide is written the same way — plain English, no hype, no jargon. $7 each, or take a bundle.

Go deeper
Get the PDF guide · $7