ElevenLabs makes synthetic speech that does not sound synthetic. That was true when it launched in 2022 and it is still the reason people pay for it, even as the platform has spread into dubbing, sound effects, music, transcription and voice agents that hold live phone conversations. The audio quality is not really in dispute. The pricing is where people get stuck. Everything runs on one pool of credits shared across products that consume them at wildly different rates, and the plan you need depends entirely on which model you generate with. Here is what each tier actually buys you, the credit arithmetic nobody explains, and the two figures the sources disagree on.
The first time you hear a good ElevenLabs read, the reaction is usually not admiration. It is mild unease. The pauses land where a person would put them, the emphasis falls on the right word, and there is none of that flattened cadence that made older text-to-speech instantly identifiable as a machine.
That is the company's whole position in the market. Cheaper synthesis exists in quantity. What ElevenLabs sells is output you can put in front of a paying audience without apologising for it.
Since 2022 it has expanded a long way past reading text aloud. There is dubbing that preserves a speaker's voice across languages, a music generator, sound effects, a transcription model, and conversational agents that answer phone calls. All of it draws from one shared credit balance, which is both elegant and the source of most confusion about what the thing costs.
I want to do the pricing properly, because it is the part every other review waves at. Then the honest limits, because there are some.
What ElevenLabs actually does
Calling it a text-to-speech tool is now about as accurate as calling Photoshop a cropping tool. Six things sit under one account, and knowing which ones you will use is the first step to picking a plan.
Text to speech
The core product. Paste a script, pick a voice, get audio. Model choice matters: Multilingual v2 and v3 for quality, Flash and Turbo for speed and half the credit cost.
Voice cloning
Instant cloning from a short sample is available from the entry paid tier. Professional cloning, which trains on longer recordings for a much closer match, starts at Creator.
Dubbing
Translate video or audio into another language while keeping the original speaker's voice. Genuinely impressive, and the most credit-hungry thing on the platform.
Speech to text
The Scribe transcription model, billed per hour rather than per character. Cheap enough that it changes the maths on podcast and interview workflows.
Music and sound effects
Generate background tracks and effects from a text prompt. Useful for filling gaps in a video edit without a stock library subscription.
Conversational voice agents
Agents that hold a live spoken conversation, for support lines and booking. Priced per minute on their own terms, separate from the credit pool.
The credits are shared. The costs are not. A minute of dubbing and a minute of Flash speech come out of the same balance at completely different rates.
That last point is the one to hold on to. A single credit pool feels simple until you spend a week's allowance dubbing a four-minute video, which is exactly how most people learn the lesson.
How credits actually work
ElevenLabs has changed its pricing model more than once. It ran on undifferentiated characters, then split quotas by model in early 2025, then simplified back to unified credits later that year. The current system is the cleanest it has been, but it still needs decoding.
One credit is roughly one character of text to speech on the top-quality models. Newer and faster models consume less, sometimes half a credit per character. Everything else on the platform spends from the same balance at its own rate.
Two consequences follow, and both are worth acting on.
First, your model choice is effectively a pricing decision. The same script run on Flash instead of Multilingual v2 costs about half as much. For a draft nobody will hear, or for content where absolute naturalism is not the point, that is free money.
Second, overage is not catastrophic but it is not cheap either. Running 150 minutes on a Creator plan means 121 included and roughly 29 billed at about $0.18, so around $27 all in. Fine occasionally. A signal to upgrade if it happens every month.
Not on your finished audio. On the eleven versions of it you generated first. Re-rendering a two-minute script six times at the highest quality while you fiddle with emphasis burns twelve minutes of allowance to ship two, and people do this constantly without noticing.
Draft on a cheap model, switch to the expensive one for the final pass. That one habit is worth more than any plan change.
Every plan, and what it actually buys
Six published tiers plus custom Enterprise. Annual billing means paying for ten months and getting twelve, roughly 17% off. Every paid plan includes a commercial licence.
I found genuine conflicts while researching this, and would rather flag them than pick one and sound confident. Starter is quoted at both $5 and $6 depending on the source and the month it was checked. Pro's allowance is reported as both 500,000 and 600,000 credits. Several aggregators also list a Business figure of $1,320 rather than $990, which looks like a seat-inclusive or annualised presentation.
The plan structure and the credit-to-minute maths are consistent everywhere, and those are what drive your decision. For the exact current numbers, check the pricing page itself, since ElevenLabs revises this more often than most.
On plan choice, the pattern across reviewers is consistent enough to trust. Creator is the default for individuals because it is the cheapest professional cloning. Pro is the inflection point for businesses, offering a better effective rate per minute than Creator. Scale matches Pro on rate, so you are buying quota and seats rather than efficiency. Business and Enterprise are volume plays.
API and pay-as-you-go rates
If a subscription is the wrong shape, the API is priced per unit. These rates were published as of September 2026 and are the clearest way to estimate a product build.
Note the spread on dubbing. A seven-fold range between the floor and the ceiling means a dubbing-heavy workflow needs testing on a real file before you promise anyone a budget.
Most people who reach for the ElevenLabs API are assembling something larger. Triggering generation from a form submission or a content pipeline is straightforward with the kind of automation we cover in our Make.com review. If the goal is a customer-facing voice agent rather than recorded audio, it is worth understanding what a full phone platform brings that a raw agent does not, which our Aircall review gets into, and how autonomous agents behave in practice, covered in our Chloe AI agent review.
Who uses it, and for what
YouTubers and faceless-channel operators
The largest visible user group by some distance. A consistent narrator voice across hundreds of videos without booking studio time, and Creator's 121 minutes a month covers a weekly upload schedule comfortably.
Course creators and e-learning teams
Course narration is where the economics are most obvious. Updating one module means regenerating one script, not rebooking a voice artist, and multilingual output turns one course into several.
Publishers dubbing into new markets
Keeping the original speaker's voice while changing the language is the feature with no real low-cost equivalent. It is also the most expensive thing per minute, so scope it before committing.
Developers shipping voice into products
IVR menus, in-app narration, accessibility readers, agents that answer the phone. Pro and above exist for this, with the 44.1 kHz PCM output and throughput limits that production use needs.
Podcasters and interviewers
Less for voice, more for Scribe. Transcription at $0.22 an hour makes show notes, clip-hunting and searchable archives trivial on any volume of back catalogue.
Once you settle on a voice, model, stability and style settings that work, write them down. Six weeks later you will be trying to match an earlier episode from memory and the settings will have drifted. A one-page note with your chosen voice ID, model and numbers saves a genuinely annoying afternoon, and RytePad's roundup of the best online notepads covers free options if you do not already keep one.
Six ways to stop wasting credits
None of these are clever. They are just the things experienced users do automatically and new ones learn by overspending.
Draft on Flash, publish on v2 or v3
Iterate on the cheap model at roughly half the credit cost, then do one final render on the quality model. Since most credit waste is re-generation rather than output, this alone can halve your consumption.
Fix the script before you generate, not after
Every typo you catch in the audio costs another full render. Read the script aloud yourself first. Awkward phrasing and missing punctuation are much cheaper to find on the page than in the waveform.
Generate in sections, not whole scripts
If one paragraph in a ten-minute read comes out wrong, you want to re-render that paragraph. Working in blocks keeps every fix small instead of forcing a full regeneration each time.
Use punctuation as direction
Commas, full stops, ellipses and paragraph breaks all shift pacing and emphasis. Rewriting the punctuation is free; re-rolling the generation hoping for a different delivery is not.
Test dubbing on one minute before you commit a project
Dubbing rates span from about $0.33 to $2.20 per minute depending on configuration. Run a single minute of real footage, look at the credits consumed, then multiply. Do not quote a client from the headline rate.
Treat consistent overage as a signal, not a cost
Overage at roughly $0.17–$0.18 per minute is fine for an unusual month. If it routinely reaches 30% or more of the next plan's price, upgrading is simply cheaper. Check your usage page monthly rather than discovering it on the invoice.
Pros, cons and honest limits
👍 What we liked
- Output quality that still leads the category on naturalness
- Professional voice cloning from $22, which is remarkable value
- Dubbing that preserves the original speaker's voice
- Six products on one account and one credit balance
- Cheap, accurate transcription at $0.22 an hour
- Clean API with transparent per-unit pricing
- Credit rollover for two months on paid plans
- A permanent free tier, not a countdown trial
- Annual billing gives two months free
👎 What to watch
- Credit model takes real effort to understand before you can budget
- Pricing structure has changed repeatedly, so old guides mislead
- Free tier has no commercial licence and requires attribution
- Dubbing costs can rise sharply with configuration
- Re-generating drafts burns allowance invisibly
- More expensive per minute than bare-bones TTS rivals
- Voice agent minutes are priced separately from credits
- Voice cloning raises consent questions you have to manage yourself
The limit worth stating plainly
Cloning a voice you do not own is the obvious ethical and legal exposure, and the platform's terms put that responsibility on you. If you clone a client's voice, get written permission covering the specific uses, and keep it. This is not hypothetical caution. It is the part of AI voice work most likely to cause a real problem, and no tool will handle it for you.
Worth paying for if…
- Delivery quality affects what the finished work is worth
- You need professional voice cloning or a consistent brand voice
- You are dubbing content into other languages
- You publish regularly and need a commercial licence
- You are building voice into a product via the API
Probably not if…
- You need plain internal reads where naturalness is irrelevant
- You generate enormous volume and cost per minute dominates
- You want one flat price with no usage arithmetic
- You only need transcription, where cheaper single-purpose tools exist
- You cannot secure consent for the voice you want to clone
Hear it before you decide anything
The free plan gives you 10,000 credits, about ten minutes of speech, with no time limit. That is more than enough to judge whether the quality justifies the price for your work. Professional voice cloning starts at $22 a month.
Free tier is permanent · commercial licence from the first paid plan · annual billing saves ~17%
Frequently asked questions
There are six published tiers. Free at $0 with 10,000 credits, Starter at around $5 to $6 with 30,000, Creator at $22 with 121,000, Pro at $99, Scale at $299 with 1,800,000, and Business at $990 with 6,000,000. Enterprise is quoted. Annual billing means paying for ten months and getting twelve, roughly 17% off.
Two figures vary between sources: Starter appears as both $5 and $6, and Pro's allowance is reported as both 500,000 and 600,000 credits. Check the pricing page for the current numbers, since ElevenLabs revises them fairly often.
Roughly 1,000 credits per minute on the quality models, Multilingual v2 and v3, or about two minutes per 1,000 credits on the faster Flash and Turbo models. So Creator's 121,000 credits is about 121 minutes of v2 speech, or closer to 240 on Flash.
Approximate is the operative word. Voice, pacing, punctuation and formatting all affect the real output, and credits are shared with dubbing, sound effects and other tools that consume them at their own rates. Use these as planning figures, not guarantees.
No. The free tier is for non-commercial use, requires attribution, and does not grant a commercial licence. Anything you publish, monetise or deliver to a client needs a paid plan, starting with Starter.
The free plan is genuinely useful as a permanent testing account rather than a trial, so use it to judge voice quality and settings. Just do not build a channel on it and hope nobody checks the licence terms.
If you are an individual creator, Creator at $22 is the answer most of the time, because it is the first tier with professional voice cloning and 121 minutes covers a regular publishing schedule. Starter only makes sense if you publish rarely and just need the commercial licence.
For businesses, Pro at $99 is the inflection point, with a better effective per-minute rate than Creator and the API quality production use needs. Scale matches Pro's rate, so you are buying quota and seats rather than efficiency. Work out your monthly minutes first, then pick the tier that covers them with a little headroom.
You are billed for overage rather than cut off, at roughly $0.18 per extra minute on Creator and about $0.17 on Pro, Scale and Business. A month where you run 150 minutes on Creator works out to around $27 total instead of $22.
The useful rule of thumb is that if overage regularly reaches 30% or more of the next plan's price, upgrading is cheaper than staying put. Unused credits on paid plans also roll over for up to two months, so your balance can reach three times your monthly quota if you have a quiet period.
Cloning your own voice is straightforward. Cloning anyone else's requires their consent, and the platform's terms place that responsibility on you rather than on ElevenLabs.
Practically, if you are cloning a client's or colleague's voice, get written permission that names the specific uses you intend, and keep the record. This is the aspect of AI voice work most likely to create a real dispute, and no amount of platform tooling substitutes for a signed agreement.
Not on its own. The conversational voice agents can genuinely hold a spoken conversation and answer calls, which covers a useful slice of front-line support and booking.
What they do not give you is the surrounding platform: numbers in multiple countries, routing rules, queues, CRM logging, call recording and team reporting. Companies that deploy agents seriously tend to run them alongside a proper cloud phone system rather than instead of one. Agent minutes are also priced separately from your credit allowance, so budget them on their own.
The verdict
ElevenLabs earns its position on output. Four years after launch it still produces the most natural synthetic speech in general use, and the expansion into dubbing, transcription, music and voice agents has been additive rather than a distraction. For anyone whose finished work is judged on how it sounds, that matters more than the price per minute.
The value at the low end is genuinely striking. Professional voice cloning and a commercial licence for $22 a month would have been a studio-budget proposition not long ago. Pro at $99 is the point where the per-minute economics turn in a business's favour, and the API rates are clear enough to plan a product build against.
What keeps it off a five is the pricing model. Credits shared across products that consume them at different rates, a model choice that silently doubles or halves your effective allowance, dubbing costs spanning a seven-fold range, and a structure that has been revised more than once in three years. None of it is dishonest. It just means you cannot buy well without reading first, and plenty of people overspend for a month or two before the logic clicks.
Start on the free tier and listen. Work out your real monthly minutes. Draft on the cheap model and publish on the expensive one. Do that and ElevenLabs is one of the highest-leverage tools a content business can put on its expense list.