Contents
- Verdict
- What is Cartesia Line?
- Best for, and not for
- Strengths and limits
- Pricing
- What people actually build with Cartesia Line
- What Cartesia Line actually costs at small-business volume
- The gotcha: overage is off by default, so calls stop at $0
- What breaks in production, and how you find out
- What it connects to that you already run
- Cartesia Line alternatives: what to pick instead, and when
- Migration and lock-in: can you get your work out?
- FAQ
Cartesia publishes a call rate of $0.03 a minute, and that number needs two additions before it means anything: $0.014 a minute for telephony on a Cartesia number, and a language model billed separately. At our reference volume of 4,000 minutes that is $120 plus $56, so $176 a month before the model. That is still inexpensive, and the managed runtime is a real saving for a developer. The thing to know before going live is that overage is off by default: when the prepaid balance reaches $0, calls stop. Worth it for a team that can write agent code and will set billing up properly. Risky for anyone who assumes a subscription keeps the phone answered.
What is Cartesia Line?
Cartesia Line is the voice agent product from Cartesia, the speech company behind the Sonic text to speech and Ink speech to text models. The documentation calls it Line and the pricing page calls it Managed Agents; they are the same thing. You prototype an agent in a no-code builder or write its reasoning in code with the open-source Line SDK, and Cartesia deploys it in a managed runtime that handles audio orchestration, scaling and observability. It is billed per minute of call duration, with telephony and the language model charged separately.
Best for
Developers who want their agent code hosted and scaled for them at a low published per-minute rate.
Not for
Buyers who need one all-in price, or who cannot have live calls stop when a prepaid balance reaches zero.
Strengths
- $0.03 a minute for call duration, a published rate the pricing page lists as reduced from $0.06
- Every plan, including Free, includes prepaid agent dollars equal to the plan price
- Agent call concurrency is generous for the price: 8 on Free, 12 on Pro, 20 on Startup and 60 on Scale
- Managed runtime deploys your agent code and handles audio, scaling, call logs and rollbacks
- The Line SDK is open source under Apache 2.0, and code can connect any language model
- Unused model credits roll over, up to twice the plan's monthly allowance
Limitations
- The $0.03 excludes both telephony and language model usage, by the pricing page's own FAQ
- Overage is off by default on paid plans, so calls stop when the prepaid balance reaches $0
- Overage is not available on the Free plan at all
- Two separate meters, model credits and agent dollars, on one subscription make the plan harder to size than it looks
- The pricing page does not state whether the call rate includes Sonic and Ink usage, or whether a number can be brought from another carrier
- Agreements for regulated data sit on Enterprise, which has no published price
Cartesia Line pricing
Cartesia sells monthly plans that bundle two things: model credits for text to speech and speech to text, and prepaid dollars for voice agents equal to the plan price. Agent calls are charged at $0.03 a minute of call duration against those dollars. Telephony is $0.014 a minute on a Cartesia-provided number and language model usage is billed separately. Overage on agent dollars is off by default and unavailable on Free.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | 20K model credits and $1 of prepaid agent usage, about 33 call minutes. 1 Cartesia phone number, 8 concurrent agent calls. No overage available |
| Pro | $5/mo | 100K model credits and $5 of prepaid agent usage, about 2 hours 47 minutes of calls. 3 phone numbers, 12 concurrent agent calls. Credit overage $65 per 1M |
| Startup | $49/mo | 1.25M model credits and $49 of prepaid agent usage, about 27 hours 13 minutes of calls. 5 phone numbers, 20 concurrent agent calls, 2 professional voice clone slots. Credit overage $45 per 1M |
| Scale | $299/mo | 8M model credits and $299 of prepaid agent usage, about 166 hours of calls. 10 phone numbers, 60 concurrent agent calls, 4 professional voice clone slots. Credit overage $38 per 1M |
| Enterprise | Custom | Custom credits, concurrency and phone numbers, with DPAs and BAAs, SSO, a shared Slack channel and security questionnaires |
| Usage rates | $0.03/min | Agent call duration on every plan, listed as reduced from $0.06. Telephony $0.014 a minute on a Cartesia-provided number. Language model usage billed separately |
Pricing reflects public plans as of November 17, 2026 and can change. Check Cartesia Line for the latest.
What people actually build with Cartesia Line
Cartesia made its name on speech models, and Line is the layer that turns those into a deployable agent. The documentation's own framing is that it brings voice to text agents you already have. That is the right way to read it: the builds that suit it start from agent logic and need a fast voice and somewhere to run.
The builds that fit
An inbound receptionist on a Cartesia-provided number. Each plan includes numbers, from one on Free to ten on Scale, so a developer can have an agent answering a real phone line without opening a carrier account. At 1,000 four-minute calls a month this is our reference workload.
A voice front end for an existing text agent. The SDK route means the reasoning is your code and can connect any language model, call external APIs and query databases. If the agent already exists as a chat product, Line gives it turn-taking, interruption handling and a voice.
A prototype that has to be demonstrated this week. The no-code Agent Builder exists for exactly this, and the Free plan's 33 minutes of agent time and one phone number are enough to put a working number in front of a client before spending anything.
Several agents for several clients from one account. An agency building for a restaurant group, a dental practice and a letting agent can run them side by side on Startup with five numbers and 20 concurrent calls, with deployments versioned and rolled back independently.
A product team that also needs the speech models directly. The same subscription's model credits pay for text to speech at one credit a character and speech to text at three credits a second, so narration, transcription and agents share a plan.
What it is genuinely bad at
Giving a non-technical buyer a single number. There is a call rate, a telephony rate, a model bill from someone else, model credits that roll over and agent dollars the page does not say roll over, and an overage switch that is off. Each is reasonable and together they are a spreadsheet. It is also a young product by the evidence available: the Line SDK repository showed 105 stars and 20 open issues on 17 November 2026, which says nothing about quality and something about how much community help exists when you get stuck.
Bottom line: Cartesia Line suits a developer who already has agent logic and wants it hosted with a fast voice; it does not suit a buyer who wants one price and no code.
What Cartesia Line actually costs at small-business volume
The pricing page, read on 17 November 2026, lists agent call duration at $0.03 a minute on every plan including Free, marked as reduced from $0.06. Its FAQ says the agent minute figures do not include telephony or language model usage. Telephony is $0.014 a minute when using a Cartesia-provided phone number, and language model usage is listed as billed separately with no rate given.
At our reference workload of 1,000 inbound calls averaging 4 minutes: call duration is 4,000 times $0.03, which is $120.00. Telephony is 4,000 times $0.014, which is $56.00. Together that is $176.00 a month, or $0.044 a minute, before the language model.
Now fit that to a plan, because you cannot buy minutes on their own. Each plan includes prepaid agent dollars equal to its price. Startup is $49 and includes $49, which at $0.03 is 1,633 minutes, matching the page's own figure of 27 hours 13 minutes. Scale is $299 and includes $299, which is 9,966 minutes. So 4,000 minutes sits awkwardly between the two. On Startup you would use the $49 and need a further $71 of agent usage as overage, for $120 in total. On Scale you would pay $299 and use $120 of it.
The cheaper route is therefore Startup with overage enabled: $49 for the plan, $71 of agent overage and $56 of telephony, which is $176. The page says usage past the balance is charged at the plan's included agent rates and billed at renewal. It does not say whether telephony draws on the same prepaid dollars or is invoiced separately; the total is the same either way, because the plan fee and the prepaid dollars are the same amount.
| Plan | Monthly price | Prepaid agent minutes at $0.03 | Agent cost at 4,000 minutes | With $56 telephony |
|---|---|---|---|---|
| Free | $0 | 33 | Not possible: no overage on Free | - |
| Pro | $5 | 166 | $5 plan plus $115 overage = $120 | $176 |
| Startup | $49 | 1,633 | $49 plan plus $71 overage = $120 | $176 |
| Scale | $299 | 9,966 | $299 plan, $120 of it used | $355 |
| Concurrent agent calls | - | - | 8 Free, 12 Pro, 20 Startup, 60 Scale | - |
| Language model | Your provider's rates | - | Billed separately | Billed separately |
The two meters on one plan
A Cartesia subscription carries two balances and it is easy to confuse them. Model credits pay for the speech models used directly: Sonic text to speech at one credit a character, which the page puts at about 750 to 800 credits a minute of audio, and Ink speech to text at three credits a second. Agent dollars pay for agent calls at $0.03 a minute.
They behave differently. Unused credits roll over, up to twice the plan's monthly credits. The page describes rollover only for credits and says nothing about agent dollars rolling over, so do not assume they do. Credit overage is priced per million credits, $65 on Pro, $45 on Startup and $38 on Scale. Agent overage is charged at the agent rate. And the page does not state whether an agent call consumes model credits for the Sonic and Ink usage inside it, or whether that is covered by the $0.03. We could not confirm it on 17 November 2026; a short test on the Free plan, watching both balances, settles it.
Bottom line: 4,000 minutes on Cartesia is $120 of call time plus $56 of telephony, so $176 a month before the language model, and Startup with overage switched on is the plan that gets you there.
The gotcha: overage is off by default, so calls stop at $0
The single most important sentence on Cartesia's pricing page is not about price. Voice agent overage is off by default on Pro, Startup and Scale, and with it off, calls stop when the balance reaches $0. An admin can enable overage with a payment method on file. On Free it is not available at all.
Think through what that means for a phone line. A business on Startup has $49 of agent usage, which is 1,633 minutes. At our reference rate of 4,000 minutes a month, about 133 minutes a day, that balance lasts roughly twelve days. On day thirteen, with the default settings, the agent stops taking calls. Nothing has broken. No error has been thrown that a customer would recognise. The phone simply is not answered by the thing you built to answer it, until the plan renews.
This is a defensible design: it protects a hobbyist from a surprise invoice, and it is the opposite of the burst billing some competitors apply. But it is the wrong default for anything customer-facing, and it is the kind of setting nobody checks because the agent worked perfectly in testing. Enable overage before go-live, and set your own alert on the balance.
The second thing to read correctly is the headline rate. $0.03 a minute is the call duration charge. With Cartesia telephony it is $0.044, and the model is still to come. The Deepgram Voice Agent API at $0.075 includes a Standard-tier model; Cartesia at $0.044 does not. Whether Cartesia works out cheaper depends on the model you attach, and for a small model on short calls it usually will, but the comparison has to be made with all three lines on the page.
Third, plan sizing is counter-intuitive. Because prepaid agent dollars equal the plan price and overage is charged at the same rate, a bigger plan does not make agent minutes cheaper. Pro with overage and Startup with overage both come to $120 of agent usage at 4,000 minutes. What a higher plan actually buys is concurrency, phone numbers, model credits and a lower credit overage rate. Choose the plan on concurrency: 12 simultaneous calls on Pro, 20 on Startup, 60 on Scale.
Bottom line: With Cartesia's default settings a live agent stops taking calls when the prepaid balance reaches $0, so enabling overage is a go-live requirement rather than a billing preference.
What breaks in production, and how you find out
The balance running out is the failure specific to this vendor, covered above. Monitor it like an uptime metric.
Concurrency is the next limit. The pricing page gives agent calls their own pool, separate from text to speech and speech to text requests: 8 concurrent calls on Free, 12 on Pro, 20 on Startup and 60 on Scale, with Enterprise custom. For most small businesses 12 or 20 simultaneous calls is ample. The risk is a campaign or an outage elsewhere that sends everyone to the phone at once. The page does not describe what a caller experiences when the pool is full, so test it deliberately before you rely on it: place more simultaneous calls than your plan allows and see what the last one hears.
The third is the part of the stack that is not Cartesia's. In the SDK route the reasoning is your code and the language model is your choice and your bill. A slow model or a slow tool call is heard by the caller as dead air, however fast the speech models are. Keep the model small for structured conversations, give tool calls a timeout and something to say while waiting, and watch the call logs for turns where the gap is long.
Cartesia does provide the tooling to see this. The documentation lists call logs for debugging conversations, evaluations with custom metrics, and deployments with version tracking and rollback. The pricing page lists evaluations as free for a limited time, so treat that as a price that may change. Use rollback as a real operational tool: deploy prompt changes as versions, and when a change makes the agent worse, go back rather than forward.
The usual voice failures still apply. Names, postcodes and reference numbers get misheard; an agent that cannot be interrupted loses callers; a handover to a human that was never tested fails on the day it matters. Test with recordings of real customers on real phone lines.
Compliance, and what we could not verify
The pricing page FAQ says Cartesia is SOC 2 Type II certified, and the Enterprise plan lists DPAs and BAAs for compliance, SSO and security questionnaires. That places the agreements a medical or dental practice would need on a plan with no published price. The page does not name HIPAA directly, and we could not read the telephony documentation, which redirected to a login page on 17 November 2026, so we cannot confirm whether a number from your own carrier can be connected. Ask both questions before committing a regulated workload.
Bottom line: On Cartesia the three things to alert on are the prepaid balance, concurrent calls against your plan's pool, and the response time of the model and tools you attached yourself.
What it connects to that you already run
Cartesia Line has two build routes and they integrate differently. The Agent Builder is for prototyping without code. The SDK is where real integration happens: the documentation says code gives you control over agent behaviour including connecting any language model, calling external APIs, querying databases and handling interruptions and turn-taking. The SDK repository is public on GitHub under the Apache 2.0 licence and is written in Python.
So the integration model is code, not connectors. Your CRM, calendar and booking system are reached by whatever your Python calls. There is no catalogue of pre-built business integrations to tick, and a business without a developer should weigh that honestly.
On speech, the pipeline is Cartesia's own: Ink for speech to text and Sonic for text to speech. That is the reason to choose the product, and it also means the voice layer is not swappable in the way it is on a bring-your-own platform. If you have cloned a voice in Cartesia, the pricing page says instant cloning is available from Pro and professional clone slots from Startup, and the agent can use it.
On telephony, plans include Cartesia-provisioned numbers at $0.014 a minute, which is the fastest route to a working line. A command-line tool deploys and tests agents from the terminal, which matters for teams that want agents in version control and a deployment pipeline rather than edited in a browser.
For everything after the call, use the pattern that keeps voice code small: have the agent post a structured summary of the call to a webhook and let an automation platform such as n8n, Make.com or Zapier update the systems of record. The agent stays focused on the conversation, and the business logic lives where a non-developer can see and change it.
Bottom line: Integration on Cartesia Line is Python you write against the open-source SDK, with Cartesia's own speech models fixed underneath and any language model on top.
Cartesia Line alternatives: what to pick instead, and when
Cartesia's position is a low call rate, a managed runtime and its own speech models. The alternatives trade one of those away for something else.
Pick the Deepgram Voice Agent API if you want the language model inside the rate and are content to host your own bridge: $0.075 a minute on a Standard-tier model, with no prepaid balance to run dry on Pay as you go. Pick Twilio ConversationRelay if you already run numbers on Twilio and want to keep them: $0.07 a minute plus call minutes, with your own model and your own server.
Pick ElevenLabs Agents if you are choosing on voice and want a subscription where exceeding concurrency is billed rather than blocked. Pick Retell AI, Bland AI or Vapi if you want a platform a non-developer can operate, with a builder, numbers and analytics. Pick LiveKit Agents if you want an open-source framework and are prepared to run the infrastructure, in which case Cartesia's Sonic can still be the voice inside it. As elsewhere on this page, the figures for the platforms we did not re-read today are from our August 2026 reviews.
| Platform | Billing unit | At 4,000 minutes | Who it suits | Does better than Cartesia Line |
|---|---|---|---|---|
| Cartesia Line | Per call minute against prepaid dollars; telephony and model extra | $176 with Cartesia telephony, before the model | Developers who want hosted agent code | A $0.03 call rate; managed runtime |
| Deepgram Voice Agent API | Per minute of connection, model included | $300 Standard, before telephony | Developers wanting one AI bill | Model in the rate; no balance that stops calls on Pay as you go |
| Twilio ConversationRelay | Per minute plus call minutes; model is yours | $315.15 local inbound, before the model | Teams already on Twilio | Uses the numbers and call flows you already have |
| ElevenLabs Agents | Subscription minutes at a flat effective rate | $320 at $0.08 a minute, model and telephony separate (August review) | Brands choosing on voice | Over-limit calls are billed, not stopped |
| Retell AI | Itemised per-minute components | About $354 on a lean configuration (August review) | Teams wanting a builder | A product a non-developer can run |
| Bland AI | One bundled per-minute rate | About $600 including the model (August review) | Buyers wanting one all-in price | Nothing left to add up |
| LiveKit Agents | Open-source framework; you pay providers | Depends on the providers you assemble | Engineers who want to own the pipeline | Swap any layer, including the voice |
Bottom line: Cartesia is the low-cost choice for a team that writes code and manages its billing settings; a business that wants one predictable all-in rate should pay more for a bundled platform.
Migration and lock-in: can you get your work out?
There are three things that can hold you on a voice platform: the code, the number and the voice. Cartesia scores differently on each.
The code is the best of the three. In the SDK route your agent's reasoning is Python in your own repository, built on an SDK published under Apache 2.0. The logic that decides what to say, which tool to call and when to hand over is yours and readable outside the platform. What you would rewrite on a move is the layer that binds that logic to Cartesia's runtime and events. An agent built only in the no-code Agent Builder is a different matter: that configuration lives in Cartesia's product, so treat the builder as a prototyping tool and move anything you intend to keep into code.
The number is the one to think about before you start. A Cartesia-provisioned number is quick, and a number your customers have learned is the hardest thing to change later. We could not confirm from public pages whether a number can be ported out or an existing number brought in. If the line matters, establish that before you print it on anything.
The voice is the durable tie, as it is with every speech company. The pipeline is Cartesia's own Sonic and Ink, and a cloned or carefully chosen voice does not exist anywhere else. Callers notice a changed voice immediately.
Commercially the exit is easy. The published plans are monthly with no annual pricing shown, so leaving is a cancellation. Unused model credits roll over only up to twice the monthly allowance, so there is no large stranded balance to lose. Keep the conversation design in a document you own and a move is a rebuild of days, not months.
Bottom line: On Cartesia Line the SDK code is yours under an open licence and the plans are monthly, so the real ties are the phone number and the voice.
Cartesia Line FAQ
How much do Cartesia voice agents cost per minute?
$0.03 a minute of call duration on every plan, listed as reduced from $0.06, as published on 17 November 2026. Telephony is $0.014 a minute on a Cartesia-provided phone number and language model usage is billed separately, so the working rate with Cartesia telephony is $0.044 a minute before the model.
Is Cartesia Line the same as Managed Agents?
Yes. Cartesia's documentation calls the product Line and its pricing page calls it Managed Agents. Both describe the same thing: voice agents built in a no-code builder or with the Line SDK and run in Cartesia's managed runtime.
What does 4,000 minutes a month cost on Cartesia?
Call duration is 4,000 times $0.03, which is $120. Telephony on a Cartesia number is 4,000 times $0.014, which is $56. Total $176 a month before the language model. On the Startup plan that is the $49 subscription, $71 of agent overage and $56 of telephony.
What happens when the prepaid agent balance runs out?
With default settings, calls stop at $0. Voice agent overage is off by default on Pro, Startup and Scale and has to be enabled by an admin with a payment method on file. It is not available on the Free plan.
Which Cartesia plan do I need for voice agents?
Agent minutes cost $0.03 on every plan, so choose on concurrency and phone numbers: 12 concurrent agent calls and 3 numbers on Pro at $5, 20 calls and 5 numbers on Startup at $49, 60 calls and 10 numbers on Scale at $299. Then enable overage for the minutes beyond the plan's prepaid dollars.
Is the language model included?
No. The pricing page lists language model usage as billed separately and gives no rate. With the SDK you can connect any language model, and you pay for it at your provider's prices.
Does the Free plan include voice agents?
Yes, in a small way: $1 of prepaid agent usage, which is about 33 minutes of calls, with 1 Cartesia phone number and 8 concurrent agent calls. There is no overage on Free, so it is for building and testing.
Do unused credits roll over?
Model credits do, up to twice the plan's monthly credits. The pricing page describes rollover only for credits and does not say that prepaid agent dollars roll over, so do not plan on it.
Is Cartesia suitable for healthcare calls?
The pricing page says Cartesia is SOC 2 Type II certified and lists DPAs and BAAs on the Enterprise plan, which has custom pricing. It does not name HIPAA directly, so confirm the agreement and its scope with Cartesia before handling patient information.
Looking at alternatives? Cartesia hosts the voice agent a developer writes and stops there: the model bill, the tools and everything the call sets in motion remain your build. Autonoly is a managed service for that downstream workflow, built around the tools a business already uses. If your team writes Python and wants a fast, inexpensive voice runtime, Cartesia fits. If nobody on the team codes, a voice runtime is the wrong first purchase. See the Autonoly review.
