Contents
- Verdict
- What is ElevenLabs Agents?
- Best for, and not for
- Strengths and limits
- Pricing
- What people actually build with ElevenLabs Agents
- What ElevenLabs Agents actually costs at small-business volume
- The gotcha: burst pricing at double rate
- What breaks in production, and how you find out
- What it connects to that you already run
- ElevenLabs Agents alternatives: what to pick instead, and when
- Migration and lock-in: can you get your work out?
- FAQ
ElevenLabs Agents is the pick when voice quality is the deciding factor, and its pricing is a clean subscription ladder with a single overage rate. As of August 2026: Free with 15 minutes, Starter $6 for 75, Creator $22 for 275, Pro $99 for 1,238, Scale $299 for 3,738 and Business $990 for 12,375, with overage at $0.08 a minute across every tier and text messages at $0.003. Concurrency rises from 4 to 40 by tier, and exceeding it triggers burst pricing at $0.16 a minute - double the standard rate. Language model and telephony costs are separate, which is the line item people forget when comparing against bundled competitors.
What is ElevenLabs Agents?
ElevenLabs Agents is the conversational agent platform from ElevenLabs, the company best known for its text-to-speech models. It bundles speech-to-text, LLM reasoning, and ElevenLabs voices into a managed real-time pipeline with turn-taking, interruption handling, telephony, and tool calling. You configure agents in a dashboard or via API, connect knowledge bases and tools, and deploy them to phone numbers, websites, or apps.
Best for
Teams that want the most natural-sounding voice output available and are willing to manage credit-based usage costs.
Not for
Cost-led buyers comparing headline per-minute rates. Once you add the separately billed language model and telephony, the effective all-in figure is higher than the subscription implies, and cheaper bundled options exist.
Strengths
- The best synthesised voice quality in the category, which is the single most noticeable thing to a caller
- A clean subscription ladder with one overage rate across every tier, so there are no surprise pricing cliffs
- Unlimited agents - the constraints are minutes and concurrency, not how many things you build
- Very cheap entry tiers at $6 and $22 a month, which makes proper evaluation affordable
- Extensive multilingual voice support, which matters for businesses serving more than one language
- Text messages at $0.003 each for follow-ups, so the agent can confirm a booking in writing cheaply
- Published concurrency limits per tier, so peak-load planning is straightforward
Limitations
- Language model and telephony are billed separately, so the subscription is not the whole bill
- Burst pricing at $0.16 a minute doubles your rate exactly when you are busiest, which is the worst moment for it
- Included minute allowances are oddly specific and easy to overrun, and overage is $0.08 rather than your blended rate
- Voice quality is the strength; conversational logic and integrations are less differentiated than specialist voice agent platforms
- Concurrency of 20 on the $99 Pro tier can be tight for a business with a genuine morning rush
- The cheapest tiers are too small to run anything real, so the practical entry point is Pro at $99
ElevenLabs Agents pricing
ElevenLabs sells bundled minute allowances on monthly subscriptions with one flat overage rate, and bills the language model and telephony separately. Published pricing as of August 2026: Free at $0 with 15 minutes and 4 concurrent calls; Starter $6 for 75 minutes and 6 concurrent; Creator $22 for 275 minutes and 10 concurrent, with a $11 first month; Pro $99 for 1,238 minutes and 20 concurrent; Scale $299 for 3,738 minutes and 30 concurrent; Business $990 for 12,375 minutes and 40 concurrent; Enterprise custom. Overage is $0.08 a minute on every tier, and burst pricing applies at $0.16 a minute - double - when you exceed your concurrency limit. Text messages are $0.003 each. There is no limit on how many agents you create; the limits are concurrency and minutes.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | 15 minutes, 4 concurrent calls (as of Aug 2026) |
| Starter | $6/mo | 75 minutes, 6 concurrent |
| Creator | $22/mo | 275 minutes, 10 concurrent; $11 first month |
| Pro | $99/mo | 1,238 minutes, 20 concurrent |
| Scale | $299/mo | 3,738 minutes, 30 concurrent |
| Business | $990/mo | 12,375 minutes, 40 concurrent |
| Overage and burst | $0.08-$0.16/min | Overage $0.08/min on all tiers; burst pricing $0.16/min when you exceed your concurrency limit. Text messages $0.003 each. LLM and telephony billed separately |
Pricing reflects public plans as of August 6, 2026 and can change. Check ElevenLabs Agents for the latest.
What people actually build with ElevenLabs Agents
ElevenLabs Agents gets chosen for one reason and it shapes every build on it: the caller cannot tell. That makes it the pick wherever the phone call is part of the brand rather than a cost centre, and it makes a specific set of workflows worth the premium.
New-patient enquiry handling for a dental practice or a private medical clinic is the clearest case. A prospective patient rings about a treatment they are nervous about, and the difference between a natural voice and an obviously synthetic one is the difference between a booking and a hang-up. The agent answers, works from a knowledge base of treatments, prices and practitioners, checks the schedule through a tool call, and books - then sends the confirmation by text at $0.003 a message rather than through a separate messaging platform.
Multilingual reception for a restaurant group or a clinic serving a mixed community is the second, and it is where ElevenLabs' voice library earns its keep. One agent, several languages, the same quality in each, without standing up a separate deployment per language.
Concierge and reservation lines for restaurants and salons are the third. These are short calls where the caller's impression forms in the first three seconds, the conversation is structured enough that a small model handles it, and the whole value is that it did not sound like a phone tree. A salon booking a regular client's usual service with their usual stylist is a ninety-second call that costs about twelve cents of ElevenLabs minutes.
Two more worth naming: a veterinary out-of-hours line whose entire job is to sound calm, triage urgency, and get a distressed owner to the emergency service quickly, with any clinical question escalated immediately; and brand-consistent outbound follow-up for a business that has already cloned a founder's or receptionist's voice elsewhere in the ElevenLabs account, where the agent simply reuses it.
What ElevenLabs Agents is genuinely bad at
Being the cheapest, and being the most operationally mature. The subscription is not the bill - the language model and telephony are billed separately on top - so a cost-led comparison against a bundled competitor flatters it. And the conversational logic, integration tooling and call-handling controls are less developed than on the specialist voice agent platforms; ElevenLabs is a synthesis company that built an agents product, and it shows in that order. There is also a quality problem nobody expects: the voice is good enough that callers often do not realise they are talking to a machine, which makes disclosure a decision you have to make deliberately rather than a default you can inherit.
Bottom line: Build on ElevenLabs Agents where the caller's first three seconds decide the outcome, and use something cheaper where the phone is a channel rather than the brand.
What ElevenLabs Agents actually costs at small-business volume
Do the division on ElevenLabs' published tiers and something jumps out that the pricing table does not advertise. Starter is $6 for 75 minutes. Creator is $22 for 275. Pro is $99 for 1,238. Scale is $299 for 3,738. Business is $990 for 12,375. Divide each by its allowance and every single one comes to $0.08 a minute - which is also, exactly, the overage rate on every tier. As of August 2026 there is no volume discount here at all. Buying a bigger plan does not make your minutes cheaper.
That is genuinely unusual and it simplifies the decision enormously. Since the rate never changes, the only thing a higher tier buys you is concurrency, which rises from 4 concurrent calls on Free to 6, 10, 20, 30 and 40 across the ladder. So the question is not how many minutes you will use - overage handles that at the same price - it is how many calls you need to handle simultaneously. Choose your tier on concurrency and let the minutes fall where they may.
On our 1,500-minute reference workload the cheapest correct answer is usually Pro at $99, which includes 1,238 minutes, with the remaining 262 at $0.08 coming to about $21 - roughly $120 a month for ElevenLabs' share. Moving up to Scale at $299 buys headroom you may not need. Only do it if 20 simultaneous calls is genuinely insufficient.
The line people forget when comparing against bundled competitors: the language model and telephony are billed separately on top, based on usage. Text-to-speech, speech-to-text, knowledge bases and RAG are included across paid plans - but the model doing the thinking and the carrier carrying the call are not. Price those from their own providers and add them, or you will under-budget against a platform quoting one all-in rate.
| Layer | Rate | Bundled or separate | At 4,000 minutes |
|---|---|---|---|
| Agent minutes, any tier | $0.08/min effective | Billed by ElevenLabs | $320.00 |
| Text-to-speech | Included | Bundled on all paid plans | $0 |
| Speech-to-text | Included | Bundled on all paid plans | $0 |
| Knowledge base and RAG | Included | Bundled on all paid plans | $0 |
| Language model | Provider's own rate | Separate, billed on usage | $12 to $640, by model |
| Telephony, US local inbound | $0.0085/min plus $1.15/mo number (Twilio) | Separate, billed by your carrier | $35.15 |
| Text messages | $0.003 each | Billed by ElevenLabs | $3 per 1,000 confirmations |
| Concurrency | 20 on Pro, 30 on Scale, 40 on Business | Included with the tier | Tier choice, not a line item |
| Burst, above your concurrency limit | $0.16/min | Billed by ElevenLabs | Double rate while over |
| Realistic monthly total | - | - | ~$370 to $470 |
The all-in stack at 1,000 calls a month
At 1,000 inbound calls averaging four minutes - 4,000 minutes - the flat rate produces a result worth stating plainly: your ElevenLabs bill is $320 whichever tier you are on. Pro at $99 plus 2,762 overage minutes at $0.08 is $319.96. Scale at $299 plus 262 overage minutes is $319.96. Business at $990 is worse only because you have prepaid for minutes you will not use. The two plans are identical on money and differ only on concurrency, 20 lines against 30.
Then add the two lines the subscription does not cover. Telephony on Twilio's published US local inbound rate is $0.0085 a minute plus $1.15 for the number, so $35.15. And the language model is yours to buy: Retell AI's published per-minute model table, the nearest public anchor, runs from $0.003 to $0.16 a minute, which is $12 to $640 at this volume. A realistic small-model configuration lands the all-in figure around $370 to $470 a month, which is competitive with Bland AI's $600 and above a lean Retell AI configuration at $354.
Bottom line: Every ElevenLabs Agents tier costs exactly $0.08 a minute, so at 4,000 minutes you pay $320 on Pro and $320 on Scale - the only thing the higher tier buys is ten more concurrent lines.
The gotcha: burst pricing at double rate
The mechanism to understand here is what happens when you exceed your concurrency limit, because it is a price event rather than an error. ElevenLabs applies burst pricing: excess calls are charged at double the standard rate, $0.16 a minute instead of $0.08, as of August 2026. Your calls still connect - genuinely better than being rejected - but they cost twice as much while you are over.
Combine that with the finding above and the tier ladder makes complete sense. Since minutes cost the same everywhere, the entire economic function of upgrading is to avoid burst pricing. If your busiest hour routinely pushes past your concurrency limit, you are paying a 100 percent premium on those minutes, and at some point the higher tier's monthly fee is cheaper than the doubling. Work out how many burst minutes it takes to cover the upgrade - between Pro and Scale, the $200 difference is covered by 2,500 burst minutes - and set an alert rather than discovering it on an invoice.
The subtler trap is that burst pricing is invisible in the moment. Nothing fails, no caller is turned away, and the only symptom is a bill larger than your minute count suggests. If your ElevenLabs invoice does not divide cleanly by $0.08, burst is the reason, and it is telling you your concurrency tier is wrong.
One more, minor but real: Creator is advertised at $11 for the first month against $22 thereafter, so the first invoice is not the recurring one.
Bottom line: If your ElevenLabs Agents invoice does not divide cleanly by $0.08 a minute, you are paying burst pricing at double rate and your concurrency tier is wrong.
What breaks in production, and how you find out
Concurrency is the operational number and it is lower than several competitors at comparable price points - 4 on Free, 6 on Starter, 10 on Creator, 20 on Pro, 30 on Scale, 40 on Business. Because exceeding it triggers billing rather than failure, the failure mode is financial rather than functional, which is a kinder design than dropped calls but means the monitoring is on you. Track concurrent call peaks, not averages, and alert as you approach the tier limit.
The reason to choose ElevenLabs is voice quality, and that shapes what goes wrong in a way worth naming. Its synthesis is the best in this category by common consent, which produces a specific and underappreciated problem: callers often do not realise they are talking to a machine. Whether your agent should disclose that it is an AI at the start of a call is a compliance and trust question in several jurisdictions and industries, and it deserves a deliberate decision rather than a default. The better the voice, the more that decision matters.
The generic voice failure modes apply and should be tested hard: latency degrading into caller and agent talking over one another, mishearing of names, postcodes and reference numbers, and an escalation path to a human that was never exercised under load. Because the language model is configured and billed separately here, latency is partly a choice you control - a heavier model gives better answers and slower ones, and past roughly a second of delay the conversation stops feeling natural however good the voice is.
Knowledge bases and RAG are included across paid plans, which is the fastest route to an agent that genuinely answers your callers' questions. The failure mode there is the one every retrieval system has: the knowledge base reflects the day you uploaded it, and an agent confidently quoting last season's opening hours does more damage than one that admits it does not know. Put re-uploading on a schedule and give it an owner.
Compliance, recording and what we could not verify
This is the honest gap on this page. ElevenLabs' Agents pricing page publishes plan prices, minutes, concurrency and overage, and says nothing about HIPAA, a BAA, data residency or retention. We tried to read its HIPAA documentation page on 2026-08-06 and the URL redirected to a documentation host that returned a 404, so we have no verified statement to report. Treat BAA availability as an enterprise sales question and get the answer in writing before any protected health information goes near a call.
The competitors are clearer on this specific point, which is worth knowing if you are a medical clinic or a dental practice: Vapi publishes HIPAA as a $2,000 a month add-on, Retell AI, Bland AI and Phonely all place a signed BAA on their Enterprise tier, and Synthflow publishes SOC 2, GDPR, HIPAA and ISO 27001 alongside its $30,000 annual floor. Call recording consent is your obligation regardless of vendor - the rules differ by state and country, and no platform's compliance page discharges it for you.
Bottom line: ElevenLabs bills rather than blocks when you exceed concurrency, so the failure is financial and silent - track peak simultaneous calls, not monthly minutes.
What it connects to that you already run
Telephony is included across paid plans as a capability while the carrier usage is billed separately, so a business phone number can be connected rather than replaced - the right architecture, and the one that protects you later. Text messaging is available at $0.003 a message, so an agent can confirm a booking by text at the end of a call without a separate messaging platform.
For business systems, the connection is tool calling and webhooks to endpoints you control, so the practical pattern matches the other developer-oriented voice platforms: ElevenLabs handles the conversation, and the automation platform you already run does the work against your CRM, calendar or booking system. There is no drag-and-drop connector catalogue between a phone call and your database, and that is the honest gap for a small business without a developer or an existing automation layer.
Knowledge bases with RAG on every paid tier cover the read-only side with no integration at all - services, pricing, policies, opening hours, the questions people actually ring about. For many small businesses that is most of the value and requires connecting nothing.
The wider ElevenLabs platform matters if you already use it. If you have built voices, cloned a founder's voice for brand consistency, or run text-to-speech elsewhere in the business, the agents product sits in the same account and the same voice library - a real convenience and one of the better reasons to choose it.
Bottom line: Knowledge bases and RAG are included on every paid ElevenLabs plan, so an agent that answers your callers' actual questions needs no integration work at all.
ElevenLabs Agents alternatives: what to pick instead, and when
The case for ElevenLabs is voice quality, so an honest alternatives section has to start by asking whether that is really what is deciding your purchase. If it is, there is no close substitute and you should stop reading. If it is not, several platforms do the rest of the job better.
Pick Retell AI if you want the same voice without the rest of the ElevenLabs product. Retell resells ElevenLabs speech at $0.040 a minute inside its own component stack, so you can have the synthesis and still pick your own model, get 20 concurrent calls included, and see every rate before you build. At 4,000 minutes a premium Retell configuration with ElevenLabs voices is $1,082 against roughly $370 to $470 all-in on ElevenLabs itself, so this is a control decision rather than a cost one.
Pick Bland AI if the separately billed language model is what is bothering you. Bland's single per-minute rate includes the model with no token charges, which at 4,000 minutes is about $600 all-in - more than ElevenLabs but with nothing left to price. Pick Vapi if you have a developer and want to hold every provider account yourself, including your ElevenLabs one.
Pick Phonely if concurrency is the real constraint. ElevenLabs charges double the rate for exceeding your limit; Phonely gives unlimited concurrent calls on every paid plan and never bursts. It is far more expensive at this volume - roughly $1,125 on Pro once overage is counted - but for a business whose calls arrive in unpredictable spikes that may be the cheaper failure mode. Synthflow is only relevant at contact-centre scale, where its $30,000 annual floor buys scoping and launch support rather than a better voice. If you are still not sure which of these your business needs, our page on what automation costs is a better starting point than a demo.
| Platform | Billing unit | At 4,000 minutes | Who it suits | Does better than ElevenLabs Agents |
|---|---|---|---|---|
| ElevenLabs Agents | Subscription minutes, flat $0.08 | ~$370-$470 all-in with a small model | Businesses where the caller's first impression is the product | Voice quality; a rate that never changes with volume |
| Retell AI | Itemised per-minute components | $354 lean, $1,082 with ElevenLabs voices | Teams that want the same voice plus full component control | Publishes every rate; 20 concurrent calls included free |
| Bland AI | One bundled per-minute rate | ~$600 on Start including transfers | Businesses that want no separate model bill | Model included in the rate; nothing left to price |
| Vapi | $0.05/min platform, providers at cost | $235 verifiable, plus 3 AI components | Teams with a developer | Your own provider keys; a published HIPAA price |
| Phonely | Subscription minutes with high overage | ~$1,125 on Pro at $0.30 overage | Unpredictable, spiky call volume | Unlimited concurrency, so no burst pricing at double rate |
| Synthflow | Enterprise annual contract | $2,500/mo floor, no per-minute rate published | Contact centres | Published SOC 2, HIPAA, GDPR and ISO 27001 posture |
Bottom line: If voice quality is not genuinely deciding the purchase, a lean Retell AI configuration does the same job for less and publishes what every component costs.
Migration and lock-in: can you get your work out?
The lock-in question here is unusual, because the thing that binds you is the thing you came for: the voice. If you have built a brand around a particular synthesised voice, or cloned a real person's voice for your agent, that asset is specific to this vendor. No competitor reproduces it, and callers who have heard your agent for a year will notice if it changes. That is a softer form of lock-in than a contract but a more durable one - and, to be fair, exactly what you are paying for.
Everything else is comparatively loose. The plans a small business would use are monthly subscriptions rather than annual contracts, so leaving is a cancellation rather than a negotiation. Because the language model and telephony are billed separately by other providers, those relationships are yours and survive a platform change - your carrier account and your model account do not belong to ElevenLabs.
The phone number is the practical asset to protect, as with every voice platform. Connect a number you control from a carrier you already have rather than taking one from the vendor, and the largest switching cost in voice disappears before it exists.
What you rebuild on a move is the agent configuration, the prompt and conversation design, the tool definitions and the knowledge base contents. Keep the knowledge base's source material in a store you own so re-uploading is trivial, and keep the conversation design - what the agent says, what it must never say, how it confirms a spelling, when it escalates - in a plain document outside the platform. That document is the genuinely valuable artefact, and with it a rebuild elsewhere is a week rather than a quarter.
Bottom line: The one thing that does not port off ElevenLabs Agents is the voice itself, which is the softest and most durable form of vendor lock-in in this category.
ElevenLabs Agents FAQ
What does ElevenLabs Agents cost in total?
Start with the subscription, then add two things. On the reference workload of 1,500 minutes a month, Pro at $99 covers 1,238 minutes and the remaining 262 bill at $0.08, so about $120. On top you pay your language model provider and your telephony separately, which typically adds a few cents a minute - realistically taking the all-in figure to somewhere around $160 to $220. Compare that against bundled competitors on all-in cost, not on the subscription alone.
What is burst pricing and how do we avoid it?
When you exceed your plan's concurrency limit, calls bill at $0.16 a minute instead of the standard rate - double. Concurrency means simultaneous calls, so Pro's limit of 20 is about your busiest moment, not your monthly total. The way to avoid it is to model your peak hour honestly: a business with a Monday-morning rush or a response to an advertising campaign can breach a limit that looks generous against monthly volume. Upgrading a tier is usually cheaper than sustained burst rates.
Is the voice quality actually worth paying for?
It depends entirely on what the call represents. For a clinic, a professional practice or a premium service where the phone is the first impression, a natural-sounding voice materially affects how callers respond, and the difference is audible within seconds. For internal call routing, order status lookups or outbound qualification, a cheaper voice is perfectly adequate and the money is better spent elsewhere. Be honest about which of those you are building.
What breaks with ElevenLabs Agents in production?
Concurrency breaches during peaks, which cost double exactly when you can least afford it. Minute overruns are the second - the allowances are specific numbers rather than round ones, and it is easy to plan against the wrong figure. Beyond pricing, the usual voice agent failures apply: latency, transcription struggling with noise and accents, and the agent not recognising a caller who needs a human. Test with real calls from real phones in real conditions.
Who should not choose it?
Cost-led buyers, because the separately billed model and telephony mean the effective rate is higher than the subscription suggests and bundled competitors are cheaper all in. Also businesses that need deep conversational logic and integration into operational systems, where specialist voice agent platforms have more mature tooling. ElevenLabs' differentiator is the voice; if the voice is not what is deciding your purchase, you may be paying for the wrong thing.
How locked in are we?
The prompts and call flows port and phone numbers can generally be moved, but the voice does not - and if voice quality is why you chose the platform, that is the switching cost that actually matters. Your callers get used to a particular voice, and changing it is noticeable. Keep your scripts and transcripts exported so a migration starts from a known position, and be realistic that a move means accepting a different sound.
How many minutes will we actually use?
Multiply expected calls by average duration, then add a margin - real calls run longer than test calls, because callers pause, repeat themselves and go off-script. A business taking 500 calls a month at a genuine three-minute average is 1,500 minutes, which sits on Pro at $99 with modest overage. If your average is closer to five minutes, you are at 2,500 and Scale at $299 becomes the better buy. Measure your current call durations before choosing a tier.
ElevenLabs or a bundled platform?
ElevenLabs when voice quality is the point and you are willing to manage the model and telephony separately. A bundled platform when you want one number on the invoice and a caller-acceptable voice is good enough. The gap in voice quality is real and shrinking; the gap in billing simplicity is real and stable. For most small businesses the deciding question is whether the phone call is part of your brand experience or just a channel.
Can you build a voice agent that sounds right?
Yes, and getting the voice and the script right together is most of the work - a good voice reading an awkward script still sounds like a robot. We scope on a free automation audit, pick the platform on whether voice quality genuinely matters for your callers, write and test the conversation with real calls rather than in a quiet room, and hand it over with prompts, number and documentation you own plus a recurring report of what it handled and what it escalated.
Looking at alternatives? If a caller hearing a natural voice matters to your business - a clinic, a premium service, anything where the first impression is the product - ElevenLabs is the strongest option and the quality difference is genuinely audible. Budget the language model and telephony separately, and watch concurrency, because burst pricing at double rate lands precisely when call volume spikes. On lock-in, prompts and numbers port but the voice does not, and if the voice is why you chose it that is real switching cost. Autonoly does not compete on voice; this is a different problem entirely. See the Autonoly review.
