Cartesia Pricing: Plans, Credits and Limits (2026)
Oct 31, 202613 min read
Cartesia plans, prices and limits in one table
Cartesia pricing tiers and what sets your bill
Cartesia prices on five tiers (Free, Pro, Startup, Scale and Enterprise), and the monthly fee is only one of five meters that set your bill, which means you can't predict your bill from your plan name alone, much as you can't judge a workflow built in Activepieces by its first step.
Credits, prepaid agent dollars, concurrency and overage rates matter just as much.
Cartesia is a voice AI platform that sells text-to-speech (TTS), speech-to-text (STT) and voice agents, all drawn from one credit pool.
A wrong number in a cost model is worse than none. Read each figure from that page, then apply the units below.
Every current plan side by side
A flat fee looks comparable across plans, but each plan meters credits, concurrency and overage differently, and those decide the real cost.
| Plan | Monthly price | Included credits | Prepaid agent dollars | TTS and STT concurrent requests | Overage per million credits |
|---|---|---|---|---|---|
| Free | Flat fee per month | Credits per month | Agent balance in dollars | Simultaneous requests allowed | Rate per million credits |
| Pro | Flat fee per month | Credits per month | Agent balance in dollars | Simultaneous requests allowed | Rate per million credits |
| Startup | Flat fee per month | Credits per month | Agent balance in dollars | Simultaneous requests allowed | Rate per million credits |
| Scale | Flat fee per month | Credits per month | Agent balance in dollars | Simultaneous requests allowed | Rate per million credits |
| Enterprise | Flat fee per month | Credits per month | Agent balance in dollars | Simultaneous requests allowed | Rate per million credits |
Prices and plan limits checked against cartesia.ai on October 31, 2026.
Concurrency is the column that bites first. If your voice workload peaks at more simultaneous calls than a tier allows, extra credits don't help, and you must move up a tier or queue calls.
What a credit buys on each model or feature
Identical call volume can cost very different amounts, because a credit isn't a fixed amount of audio. Each model or feature consumes credits at its own rate, so the bill depends on whether you run TTS, STT or an agent.
Identical call volume can cost very different amounts, because a credit isn't a fixed amount of audio.
If you are running this arithmetic for your own team, see what the same workload costs on Activepieces.
Monthly bill for a 5,000-minute voice agent
The workload and its assumptions
I assume flat volume and one agent rate, so a month with more calls will cost proportionally more.
Cost on each plan, step by step
On Free there's no plan fee, so all $300 comes from usage.
On Pro, Cartesia's $5 monthly plan fee is the smaller part of the bill. Tradeswithai.com found the overage of $295.02 is 59 times larger, which means the headline price tells you almost nothing about what you'll pay. Together they make $300.02.
(chart)
Where the cheapest plan stops being cheapest
At 5,000 minutes, Free is $0.04 cheaper than Pro ($299.98 against $300.02). Price alone gives no reason to upgrade.
Upgrading would pay off only if Pro's inclusions reduced the overage by more than its $5 fee, and at this rate that would mean a meaningful cut to the 5,000 metered minutes, meaning Free is the rational choice.
If it doesn't, the gap stays within pennies.
Turn the decision on limits instead, such as how many calls your agent must handle at once. Check your peak simultaneous calls against each plan's concurrency cap before you commit.
Because usage multiplies the rate, a 20% rise in volume adds about $60 to either bill, while the plan fee stays at $5.
Hidden costs and limits in Cartesia billing
Concurrency caps, credits that expire, a dearer overage rate and per-character metering all push a Cartesia bill above the plan fee.
In a voice agent's minute, HireFinn puts telephony at 35.4% of cost, text to speech at 25.8%, speech to text at 20%, LLM prompt at 16.7% and LLM output at 2.1%.
So the metered text-to-speech line you price from is only about a quarter of each minute.
The vendor figures that move the total:
- Concurrent TTS caps of 2, 3 and 5 across the tiers
- Credits that roll over up to 2x your monthly credits
- Pro overage at $65 per million versus $45 per million on Startup
- Voice cloning priced outside the base plan
- Speech-to-text billed separately
- Terms that appear only on enterprise quotes
At 5 concurrent TTS streams, a sixth caller during a busy hour gets queued or rejected. A cap of 2, 3 or 5 is a ceiling on simultaneous callers, not on monthly volume, so peak traffic, not average, decides your tier.
A quiet month is not lost, because unused credits roll over up to 2x your monthly credits. A busy month tips you into overage, and, so exhausting credits early makes the last stretch of the month the dearest.
Before the first call, a branded voice adds a line item, since voice cloning is priced apart from the base plan. Speech-to-text is also billed separately. At 20% of the minute, leaving it out of your estimate understates the bill by a fifth.
If your caps or volumes land on Enterprise, you can't model the terms from the pricing page, because they're quoted, not published. The number that matters is the contract, and telephony at 35.4% of the minute will sit outside it either way.
Cartesia against ElevenLabs, Deepgram and OpenAI voice pricing
Every figure comes from that vendor's own pricing page, checked in October 2026.
Comparison table on four dimensions
Read the 5,000-minute column first, because it's the only one that reflects what you'd pay.
Telephony is the line the table leaves out. Cartesia's rate is $0.014 per minute, identical to Twilio's outbound rate, leaving features as the true differentiator. Twilio inbound is $0.0085 per minute, making it roughly 39% cheaper than outbound.
Telnyx charges $0.005 outbound and $0.0032 inbound, with the latter roughly 36% cheaper.
When Cartesia is the cheaper choice
Cartesia wins when the table shows a lower 5,000-minute cost than the others and its concurrency cap sits above your peak call count.
Check the cap first.
On telephony, Cartesia matches Twilio outbound. At 5,000 outbound minutes, that's $70 on either, so Cartesia gives you no saving there and no penalty either.
When a competitor is the better buy
Pick a competitor when its free tier covers your testing or its concurrency limit fits your peak without a tier jump. Both can outweigh a small gap in entry price.
Telephony is where the gap is largest. At 5,000 outbound minutes, Telnyx's $0.005 comes to $25, against $70 at $0.014. That's $45 a month saved, about 64% off, with no change to the voice vendor.
Inbound shows the same pattern: Twilio's $0.0085 comes to $42.50 and Telnyx's $0.0032 to $16, so Telnyx is roughly 62% cheaper than Twilio.
If your workload is mostly inbound, the Telnyx saving is $26.50 at the same volume. Price the telephony carrier separately from the voice vendor, because the cheaper carrier is a separate decision from the cheaper voice plan.
Worth checking against a plan that does not meter every step: one credit covers a whole run on Activepieces.
Connecting Cartesia to business workflows with Activepieces
Don't expect Activepieces to make Cartesia cheaper: it's an automation builder with 746 integrations and an MIT-licensed core, and if an integration you need doesn't exist, you can generate one with AI and add it privately to your own instance, while a run costs 1 credit however many steps it has, so step count doesn't affect the per-run cost.
When a workflow tool is worth adding
Add it only if speech must be generated in response to events in other tools: a form submission, a new spreadsheet row, a support ticket. Each of those needs something to notice the event and send a text-to-speech request.
If your own application already calls Cartesia directly, a workflow tool adds a second system to maintain and changes nothing about the bill. Skip it.
In the screenshot, a workflow titled "CV Scanning" has five steps, which is the kind of flow where it fits. A Web Form trigger labeled "Human Input" starts it.
Then Google Sheets logs a row, a step extracts text from a PDF, a text AI step summarizes it, and a second Sheets step writes the result back.
With the Web Form step open, you can see that the trigger gives you a published form URL and a draft one, plus a warning about "Wait for Response". Nothing in this flow produces audio.
Replace the summary step with an HTTP request to Cartesia and the same shape would turn each submission into a spoken summary. That HTTP step is where credits start being spent, so it's the step to price.
The number of flows around it is not a cost: the Activepieces pricing page lists "unlimited flows" on every tier from Free ($0, 1,000 credits a month) through Ultimate, whereas Zapier's pricing page bills a task per action and Make's bills a credit per module.
Break the fragile mega-flow into ten small ones and the only meter left running is Cartesia's credits per request.
Keeping credit use predictable
Every trigger fire becomes a paid request unless you stop it. A spreadsheet that gets bulk-pasted, or a form that gets spammed, could send hundreds of speech requests you never intended, and each draws on your credit balance.
On the Activepieces side that burst is cheap to absorb: a run is 1 credit however many steps it has, against the Free plan's 1,000 credits a month, and if those run out the flows stay on while runs wait for the credits to refresh.
The meter to watch is the Cartesia request each run makes.
Three habits keep the spend readable:
- Put a filter step before the HTTP call so only rows that need audio reach it.
- Cap concurrent runs below your plan's concurrency limit.
- Have the flow write each request's character count to a sheet, so you can compare real credit use against your estimate.
Monday checklist for choosing a Cartesia plan
Those habits only hold if you start from your own usage, so run these steps in order before you pick a tier:
- Measure your monthly characters and your peak simultaneous calls.
- Run the workload on the lowest tier.
- Compute the bill against the plan table, using your measured numbers.
- Set usage alerts on credits and concurrency.
- Write down the conditions that trigger an upgrade.
Measuring Cartesia characters and peak concurrency
Count the unit being metered, which is characters of synthesized speech, not minutes of calls. Pull a month of transcripts, or sample a representative week and scale it.
Then find the busiest moment, not the average: peak simultaneous calls is what hits a concurrency cap, and an average hides it. A workload with modest monthly volume but a sharp lunchtime spike can need a higher tier than a heavier, evenly spread one.
Test on the lowest tier first
Run real traffic through the cheapest plan with the model and features you actually intend to ship. Credits can drain at different rates depending on which model or feature you call, so a test on a default setting will understate your bill.
Record credits burned per thousand characters and any calls that queue or fail at the cap.
With those measured rates, do the arithmetic against the plan table: monthly characters times your observed credit cost, plus any overage, compared across tiers.
The cheapest tier that stays under both your credit use and your concurrency peak is the answer, even if a larger plan looks better value per credit.
Setting a Cartesia plan upgrade trigger
Turn on usage alerts for credits and for concurrency, set well before the limit so you have time to act.
Then write the upgrade rule down in advance: for example, upgrade when projected overage cost passes the price gap to the next tier, or when rejected calls at the cap become a recurring event.
Frequently asked questions
Most of the small questions come down to three facts, each of which changes a decision you may already have made:
- Voice cloning needs a short sample, about 10 seconds for an instant clone., so you can test a clone from a single recorded line but should supply the longer sample if you are on a newer model and want it closer to the speaker.
- First audio arrives in under 90ms, so the model is rarely the slow step in a voice agent's response; your speech-to-text, language model and network hops will cost you more.
- Yearly billing is about 20% cheaper than monthly, so on a plan you know you'll keep, commit annually, but only after your workload estimate holds up for a full year.
The four questions below cover what the body of the article leaves out.
Does Cartesia have a free plan? Yes, Cartesia lists a free tier with a small credit allowance.
Treat it as a way to test voices and measure how many credits your own scripts use. Its concurrency limit isn't built for live traffic, so a production agent will hit the ceiling quickly.
Can I change or cancel a Cartesia plan mid-month? Plans can be changed from the billing settings, but check how the page treats proration and unused credits before you do.
If a launch needs more concurrent calls, upgrade before it starts. Moving up after callers get rejected costs you the calls you lost.
Does Cartesia offer discounts for startups or annual billing? Annual billing is the discount you can count on, and it's the one in the list above.
Startup pricing is published at $49 per month, with 1.25M credits; only Enterprise pricing is custom. Don't build a budget on a discount you haven't been quoted in writing.
Is Cartesia priced per character, per minute or per credit? Per credit.
Each plan includes a pool of credits, and the model and features you use decide how fast it drains. Convert your expected characters or call minutes into credits for your chosen model, then compare that figure to the plan's allowance.
Related reading
References
Running the numbers
See what the same workload costs here.
Free forever plan, and every paid plan self-hosts at no extra cost.
See pricing Talk to sales
