ElevenLabs API Key (2026): Getting One, Costs, Limits

An ElevenLabs API key is issued inside the app, so the questions worth answering first are which product you are keying into and how the quota is counted.

The two-minute versionVoice is the strongest thing on this site and the customer list backing it is not small. Get a key if speech is the product: narration, dubbing, agents that talk. Think twice if audio is one step inside a pipeline that also needs images and video, because you will be managing three vendors and three quotas to ship one feature, and something like Synexa collapses the generation side into a single endpoint. The caveat nobody reads in time is that quota is usually counted in characters or seconds, not in requests, so your invoice tracks script length rather than call volume.

Try one run → Official site

Three products wearing one logo

The front page splits the company into ElevenCreative for content production, ElevenAgents for customer experience, and ElevenAPI for developers, with a capability row underneath covering an AI voice generator, text to speech, music, speech to text, voice cloning and dubbing. That structure matters when you go looking for a key, because the thing you are integrating determines which part of the app you end up in and which quota you consume. Somebody who wants a narrator and somebody who wants a phone agent are buying different products from the same account. Work out which of the three describes your project before you read a single price, otherwise you will be comparing figures that were never meant to be compared.

Where an ElevenLabs API key actually comes from

Not from the marketing pages. The home page offers exactly two doors, a sign-up link and a contact sales link, plus a pricing page in the navigation. Everything to do with credentials lives behind the sign-up, in the application itself, which is standard for this category and worth saying plainly because search results are full of tutorials describing menus that have since been rearranged. Create the account first, find the developer section of the app, and read the current documentation from there. If a tutorial tells you to click something that does not exist, believe the app and not the tutorial. The same goes for every screenshot on this page's competitors, including screenshots taken last month.

The quota question the home page skips

The front page sells capability and credibility, listing customers including Twilio, Disney, Stripe, Cisco, Nvidia, Meta, IBM, Klarna and Deutsche Telekom. What it does not do is put a number next to any of it, and the number you need is not requests per month. Speech services typically meter by characters synthesised or minutes produced, which changes how you forecast entirely: one long audiobook chapter can cost more than a thousand short notification lines. Before committing, find the metering unit on the official pricing page, then measure a representative script of your own and multiply. Estimating from request counts is the mistake that turns a comfortable plan into an overage two weeks after launch.

Cloning and dubbing bring a consent problem

Voice cloning and dubbing are both listed as capabilities, and both carry obligations that have nothing to do with the API. Cloning a voice means you need a defensible right to that voice, from a signed release for a real person to a licence for a character. Dubbing a video into six languages means the translation carries your brand into markets where nobody on your team can read the output. Neither is a reason to avoid the technology. Both are reasons to write down who approved each voice and each language before the files go out, because the failure mode here is reputational rather than technical, and no error log will warn you about it.

Audio as one step in a longer pipeline

Plenty of projects need narration only because a video needs narration. If the same pipeline also generates the images and the footage, you are signing up for three integrations, three sets of credentials and three quotas that run out at different times. Synexa takes the other approach: one REST endpoint with a Python SDK across FLUX, video and audio models, billed per run. It will not match a dedicated voice platform on voice-specific tooling such as a voice library or dubbing workflows, and pretending otherwise would be silly. What it does is reduce the number of vendors in a generation pipeline to one, which is the right trade when audio is a step rather than the product.

Before you sign up

Pick the product first

The site separates creative production, agents and the developer API. Your key belongs to one of those contexts, and the plan that makes sense follows from that choice rather than the other way round.

Keys live in the app

The public pages offer sign-up, contact sales and pricing. Credential management sits behind the login, so create the account before following any tutorial that promises a shortcut.

Metering is not per request

Speech services generally bill by the amount of audio produced. Measure a real script, not a test phrase, then check the unit on the official pricing page before choosing a tier.

Consent is part of the build

Voice cloning and dubbing are on the capability list. Record who authorised each voice and each language, because that paperwork is the part an incident will turn on.

Dedicated voice platform or one generation endpoint

FeatureElevenLabsSynexa
FocusVoice, speech and audio as the core productImage, video and audio model runs through one API
Listed capabilitiesText to speech, speech to text, music, cloning, dubbingFLUX, video and audio models
Voice libraryBrowsable voices organised by use caseNot a voice library product
InterfaceApps plus a developer APIREST endpoint plus a Python SDK
Billing shapePlans on the official pricing pagePay per run
Fits best whenSpeech is the deliverableAudio is one step beside images and video

From sign-up to first synthesised line

  1. Create the account
    Use the sign-up link on the official site rather than a tutorial's deep link. Layouts change, and starting from the front door means you land on whatever the current application looks like.
  2. Find the metering unit
    Open the pricing page and note exactly what is being counted before choosing a plan. Characters, minutes and requests forecast very differently once real scripts arrive.
  3. Synthesise a real script
    Run one genuine piece of copy, not a test phrase, and listen to the whole thing. Pronunciation of names, numbers and acronyms is where voice systems embarrass you, and it never shows up in a demo.
  4. Store the key properly
    Environment variable or secrets manager, one key per project, nothing committed and nothing in a client bundle. Rotate it the day anyone with access to it leaves the team.

FAQ

Is an ElevenLabs API key free to get?

The site offers a sign-up path and a separate pricing page, and does not publish plan details on the home page itself. Whether a free allowance exists, and how large it is, belongs on that pricing page, which is the only version worth quoting since these allowances change without announcement.

Where do I find the key after signing up?

Inside the application, in its developer or settings area. The public marketing pages link only to sign-up, contact sales and pricing. If an article describes a specific menu path, treat it as a hint rather than instructions, because those paths get rearranged between releases.

How is usage counted?

Speech platforms typically meter the audio produced rather than the number of calls, which means a handful of long scripts can outweigh thousands of short ones. Confirm the unit on the official pricing page and then measure one of your own real scripts against it before you pick a tier.

Can I clone a voice for commercial work?

Voice cloning is listed as a capability, and the technical part is the easy half. The rights to the voice are the real question, and they sit with you: a signed release from the speaker, or a licence you can produce later. Get that in writing before anything is published.

Which companies use it?

The home page names customers including Twilio, The Walt Disney Studios, Stripe, Telus, Cisco, Nvidia, Revolut, Meta, Zoom, Ramp, IBM, Deutsche Telekom, Klarna and Harvey. A logo wall proves the vendor can serve large accounts; it says nothing about whether the plan you are considering fits your workload.

What if I also need images and video?

Then count your vendors before you commit. A service such as Synexa runs FLUX, video and audio models behind one REST endpoint with per-run billing, which means one credential and one meter for the whole generation step instead of three that expire on different days.

One credential for the whole generation step

Synexa puts FLUX, video and audio models behind a single REST endpoint with a Python SDK and per-run pricing, so a pipeline that needs pictures, footage and sound needs one integration rather than three.

Try one run →