Cartesia Review 2026

Published September 1, 2026 · Every fact below was read from Cartesia’s own pricing page, developer documentation, changelog, Terms of Service and Acceptable Use Policy on September 1, 2026, with the source linked beside the claim.

Quick verdict

Every price, licensing term and feature claim below was read from Cartesia’s official pages on September 1, 2026, with the source linked beside the fact. Our standardized audio benchmark is designed but not yet running, so this review contains no listening-test claims, no naturalness verdicts and no scores. Cartesia markets low latency; we do not convert a marketing claim into a measurement.

Strongest advantages:

• A commercial-use licence and instant voice cloning together at $5/month — the cheapest such combination in our coverage.
• An unusually honest deprecation practice: sunsets are published in a dated table, with the error code you will receive, before they happen.
• Clones survive migration: “Instant and Professional Voice Clones work on Sonic 3.6 with the same voice IDs — nothing to recreate” (changelog, checked September 1, 2026).
• Unlimited workspace seats on every plan, per the pricing page’s own hero copy.
• Cartesia states plainly that it does not claim ownership of your inputs or outputs (§5.3).

Important disadvantages:

• Two model aliases and one snapshot stop working after October 20, 2026, at alias level.
• The Voice Changer endpoints’ sunset date has already passed with no documented successor.
• Commercial use is prohibited by default and the Terms never name which tier unlocks it.
• The Terms of Service and privacy policy are dated June 14, 2024 — over two years old, with no model names and no lifecycle language.
• Inputs and outputs feed model training by default under a perpetual, irrevocable, sublicensable licence.
• Cartesia expressly does not warrant that you own the output, and offers no IP indemnity.
• Liability is capped at the greater of six months of fees or $100.
• No published requests-per-second rate limit and no SLA.

On this page

The deprecation calendar

This section comes before the buying advice because for anyone already running Cartesia in production it is the only part that is urgent.

The October 20 rows are corroborated on three official pages. The API-changes table lists them under “Upcoming sunsets”; the older-models page states “sonic-2 and all sonic-2-* snapshots will stop working after October 20, 2026. Migrate to Sonic 3.6.” with the identical sentence for sonic-turbo; and the changelog’s July 2026 entry says “Sonic-2, Sonic-turbo, and Sonic-3-2025-10-27 will be sunsetted after October 20, 2026. Migrate to the newest stable TTS snapshot before then.”

The scope detail decides how much work you have. For sonic-2 and sonic-turbo the sunset is at alias level and covers all snapshots — the usual defensive move of pinning a dated snapshot does not save you here. For sonic-3, only the one dated snapshot is listed.

Migration is also not automatic. The August 2026 changelog states: “For customers currently using sonic-3.5, you will have to manually repoint production traffic to sonic-3.6-2026-08-27 or sonic-3.6.” The reassurance beside it is worth having: “Your voice clones carry over. Instant and Professional Voice Clones work on Sonic 3.6 with the same voice IDs — nothing to recreate”, and “Keep your existing voice IDs, request shape, and prompting — no code changes required.”

Requests to a retired model do not fall back silently. Cartesia documents the failure: “Requests to sunsetted models will return an error”, with the body carrying "error_code": "model_sunsetted".

The one that already happened

The API-changes page introduces its breaking-changes table with: “These endpoints are being sunset. Requests after an endpoint’s sunset date return an error.” Two rows in that table are POST /voice-changer/bytes and POST /voice-changer/sse, both dated August 20, 2026, both with an em-dash rather than a replacement.

That date passed twelve days before we checked, and the marketing site still publishes a Voice Changer capability page and links it from the footer. We are not going to guess whether the endpoints are actually off, whether the date moved, or whether the marketing page is simply stale — all three are possible and Cartesia does not say. What we will say is that anyone with voice conversion in production should treat this as a live breakage with no documented successor, and confirm directly.

Who should use it — and who should look elsewhere

Choose Cartesia if:

• You want commercial rights and cloning at the lowest possible price. The Pro plan is $5/month and its bullets include both “Commercial use license” and “Instant voice cloning” (pricing, checked September 1, 2026). Nothing else in our coverage pairs those two that cheaply.
• You are building real-time voice agents and want one vendor for speech in and speech out. Cartesia sells Sonic for synthesis and Ink for transcription on the same account.
• You will actually read the changelog. Cartesia’s deprecation practice is transparent and dated. That is a genuine virtue if you track it, and a liability if you do not.
• Seat count matters. The pricing page advertises “unlimited workspace seats on every plan”, including Free.

Look elsewhere if:

• You need a model that will still be there in two years. Cartesia has executed sunsets in June and August 2026 and has more scheduled for October. If stability outranks price, Amazon Polly publishes no deprecation for its standard engine at all (pricing, checked September 1, 2026).
• Your legal review needs a current contract. Cartesia’s Terms are dated June 14, 2024 and name no model, no version and no lifecycle commitment.
• You cannot have your inputs and outputs used for training. It is on by default on every plan, and the opt-out is a form rather than a setting.
• You need an SLA or a published rate limit. Neither exists on the pages we read.

Current price, free plan, commercial rights

The short version: five plans, monthly billing only, credits rather than characters, and a free tier that deliberately excludes the two things most buyers need.

Quoted from cartesia.ai/pricing, checked September 1, 2026. Minute figures are Cartesia’s own approximations for Sonic 3.6 in its comparison table, not our conversions. Billing is monthly only — the page carries no annual toggle and no yearly price. Each plan also carries a matching prepaid agent allowance ($1 on Free, $5 on Pro, and so on).

What this means for you: the free tier is a trial surface rather than a usable free product, because it carries neither a commercial licence nor cloning. The real entry point is $5, and at that price it is the most generous commercial starting position in this comparison.

Commercial rights: prohibited by default, unlocked by a tier the contract never names

The Terms of Service, clause 4.1, state verbatim:

“We hereby permit you to use the Services for your personal, non-commercial use only (unless commercial use is expressly permitted by your subscription tier)…”

The parenthesis is doing all the work, and the contract never says which tier qualifies. The only official statement connecting that permission to a price is a feature bullet — “Commercial use license” — on the Pro card of the pricing page. No Cartesia page defines the scope of that licence: what it covers, whether it survives cancellation, or what happens to work produced while subscribed.

This is a weaker position than it first appears. A pricing-card bullet is not a licence grant, and the Free tier’s silence means free output is squarely inside the default prohibition. If commercial output matters, be on a paid tier and keep the record of it.

What this review covers

This review covers Cartesia’s Sonic text-to-speech models and their published lifecycle, the Ink speech-to-text models where they bear on a voice purchase, the instant and professional cloning tiers and their gates, the published plans and credit allowances, and the contract terms governing commercial use, ownership, training and liability. The contracting entity is Cartesia AI, Inc., at 555 De Haro Street, Suite 300, San Francisco, California, per Terms §9.7.

It does not evaluate how any Cartesia voice sounds. We created no account, made no API call and generated no audio. Rows a review of ours normally carries once our benchmark runs — Tested, Model tested, Voice tested, Benchmark cost — are absent rather than filled with invented values.

One dating note governs every legal citation here: the Terms of Service and privacy policy both display “June 14, 2024”. Only the Acceptable Use Policy has been revised since, on July 23, 2025. Every lifecycle commitment quoted in this review is documentation-level, not contractual — the contract contains no model names and no sunset language at all.

Benchmark audio status

Samples are pending. Our standardized audio benchmark is designed but not running, so this page publishes no audio and no empty player (methodology). What we verified instead is documentation: the sunset tables and their scopes, plan contents and gates, concurrency ceilings, cloning sample requirements, and every contract clause quoted below, all read on September 1, 2026.

Cartesia markets latency heavily. We publish no latency figure, because we measured none and we do not restate a marketing claim as a finding.

Assessment by dimension

No numeric scores appear below. The judgments are qualitative and drawn from what Cartesia documents.

Voice quality and naturalness

We will not tell you how Cartesia sounds. No listening test has been run for this review.

What documentation settles is the envelope: which model your account can call, for how long it will exist, and what concurrency you get. On this provider that envelope changes more often than most, which is why the deprecation calendar is the section we would read first.

Voice cloning and consent

Cloning is tiered by sample length and plan. The documentation gives both: instant cloning takes a 10-second sample and requires “Pro plan or above”; professional cloning takes 30 minutes of audio and requires “Startup plan or above” (cloning guide, checked September 1, 2026). The pricing comparison table caps professional clones at 2 on Startup and 4 on Scale. Attempting a gated feature returns the API error plan_upgrade_required.

On consent, there is a gap between the policies and the product. The Acceptable Use Policy, clause 2, states verbatim: “You may only submit your own voice and audio recordings or those of others with explicit consent, and are solely responsible for ensuring that you have all of the necessary rights and consents for the Input that you provide when using our Services.” The Terms, clause 4.2, separately prohibit making available “audio of a voice of a deceased person, a candidate for a political office, or any other person’s voice without that person’s express permission”.

Those are warranties you give, not gates the product enforces. The Clone Voice API reference contains no occurrence of the word “consent” and has no attestation field — the request takes a clip, a name, a language and descriptive metadata. Cartesia places the obligation on you and does not check it.

That is a defensible design and a common one, but it is materially different from providers that require a recorded consent phrase before a clone can be created. If your compliance position depends on the platform enforcing consent, it does not here.

API, concurrency, limits

Concurrency, not requests per second, is the limit that will bite. Cartesia documents: “Your account is subject to two types of rate limits: WebSocket limits and generation concurrency limits”, with text-to-speech concurrency of 2 on Free, 3 on Pro, 5 on Startup and 15 on Scale (concurrency docs and the pricing comparison table, both checked September 1, 2026).

No requests-per-second or requests-per-minute figure is published, and no service-level agreement appears on any page we read.

Two further constraints worth knowing before you design around them. Batch transcription cannot use the current speech-to-text model: the endpoint comparison lists /stt (batch) as supporting ink-whisper only, while the current ink-2 is available on the turn-based WebSocket endpoint. And agent calls are recorded by default: “By default, Cartesia records the audio and transcript of every agent call so you can review them after the call ends”, with no per-call switch documented to turn it off mid-call.

Zero Data Retention exists but is enterprise-only and does not cover everything: the documentation scopes it to the text-to-speech and speech-to-text APIs for Enterprise customers, and excludes voice cloning.

Generation speed and latency

Cartesia positions itself on real-time performance. We publish no latency figure for it, because we have measured none and we do not restate marketing adjectives as numbers.

This section will carry measured figures when our benchmark runs.

Pricing and normalized cost

What you pay, and what you get

Cartesia bills in credits, with the pricing page providing its own credits-to-minutes approximations for Sonic 3.6. Using those figures: Pro at $5 for roughly 133 minutes works out near $0.038 per minute; Startup at $49 for roughly 1,667 minutes near $0.029 per minute; Scale at $299 for roughly 10,667 minutes near $0.028 per minute. Those are our computations from Cartesia’s published approximations, not vendor statements, and they assume you consume the whole allowance.

The important limitation: the cheap tier is cheap because it is small

The per-minute rate barely improves as you climb — roughly $0.038 to $0.028 across a sixty-fold jump in volume. What the higher plans actually buy is headroom and concurrency, not a materially better rate. Choose the tier by the concurrency and cloning you need, not by an expectation of volume discount.

Who gets the best value

• Side projects and prototypes that need commercial rights: excellent. $5/month is the lowest commercial entry in our coverage.
• Voice-agent builders: good, given prepaid agent allowances and speech-to-text on the same account.
• Long-horizon production systems: weak, because the model lifecycle demands ongoing migration work that has a real engineering cost.

Commercial usage, licensing, privacy

Who owns what you generate

Cartesia disclaims ownership plainly, in clause 5.3: “Cartesia does not claim ownership of any of your Inputs to the Services (‘Your Inputs’) or any of the Outputs you create with the Services (‘Your Outputs’).”

Read that beside clause 7.1, which is displayed in capitals in the original: “THE CARTESIA ENTITIES DO NOT REPRESENT OR WARRANT THAT YOU ARE THE LEGAL OWNER OF ANY OUTPUT…”

Those two are not contradictory, and the distinction matters: Cartesia is not claiming your output, but it is also not promising you own it. Indemnity runs one way only, from customer to Cartesia (§7.3), and liability is capped by §7.2 at the greater of total fees paid in the previous six months or $100. On a $5/month plan that ceiling is $100.

Your inputs and outputs train the models by default

Clause 5.3(c) grants Cartesia, verbatim, “a non-exclusive, irrevocable, perpetual, worldwide, royalty-free, fully paid, transferable, sublicensable right and license to use any Inputs and Outputs made available by you or otherwise generated in connection with your use of the Services at any point…”

Every word of that is load-bearing: irrevocable, perpetual, transferable and sublicensable. The opt-out under clause 5.3(d) is a manual form rather than an account setting, it is scoped by category, and it is forward-only — the clause states it “will not affect any uses of Your Content (or improvements to the Models as a result) prior to that date.”

No dashboard toggle is documented on any plan. If this matters to you, submit the form before you send anything sensitive, not after.

Pros

• Commercial licence plus instant voice cloning at $5/month — the cheapest such pairing in our coverage.
• Dated, published deprecation tables with named replacements and a documented error code.
• Voice clones survive model migration with the same voice IDs and no code changes, per the changelog.
• Unlimited workspace seats on every plan, including the free one.
• Speech-to-text and text-to-speech on one account, useful for voice agents.
• Cartesia disclaims ownership of your inputs and outputs in plain terms.
• Professional cloning at $49/month is well below the enterprise gates several competitors impose.

Cons

• Two model aliases stop working after October 20, 2026, and pinning a snapshot does not help.
• The Voice Changer endpoints’ sunset date has passed with no documented replacement, while the marketing page remains live.
• Commercial use is prohibited by default; the enabling tier is named only on a pricing card, never in the contract.
• Terms of Service and privacy policy are dated June 14, 2024.
• Perpetual, irrevocable, sublicensable training licence over inputs and outputs, on by default.
• Opt-out is a manual form, category-scoped and forward-only.
• No warranty that you own the output, and no IP indemnity.
• Liability capped at the greater of six months of fees or $100.
• No published rate limit and no SLA.
• Agent calls recorded by default with no documented mid-call off switch; zero-retention is enterprise-only and excludes cloning.

Best use cases

And the inverse, because a recommendation without one is not worth much: if nobody on your team will be watching a vendor changelog in six months, this is not the provider to build on — not because it hides its changes, but because it makes so many of them.

Alternatives

Organized by the reason you would leave.

• You need model stability over years. Amazon Polly publishes no deprecation for its standard engine and bills per character with no subscription (pricing, checked September 1, 2026).
• You need cloning with platform-enforced consent. Speechify’s API requires the speaker to record a phrase it issues, with no unattended path (consent docs, checked September 1, 2026).
• You want the deepest documented cloning at a consumer price. ElevenLabs documents instant and professional cloning from $6/month (pricing, checked September 1, 2026).
• You want to self-host and avoid lifecycle risk entirely. Resemble AI publishes MIT-licensed models with an express commercial grant.
• You want the whole market in one view. The ranking of ten providers compares pricing, free tiers, licensing and cloning side by side.

Direct comparison links

Head-to-head pages pairing Cartesia against individual competitors are on this site’s roadmap but not published yet, and we do not link to pages that do not exist. Until they are live, the flagship ranking carries side-by-side pricing, licensing and feature tables for all ten providers we cover.

Final verdict

Choose Cartesia when budget is the binding constraint and someone on your team will keep an eye on the changelog. Five dollars a month for a commercial licence and instant cloning is genuinely the best entry price in this comparison, professional cloning at $49 undercuts the enterprise gates elsewhere, and the deprecation practice — dated tables, named replacements, a documented error code — is more honest than most vendors manage.

Do not choose it when you need to build once and leave it running. Cartesia retired a model generation in June, retired two endpoints in August with no successor named, and will retire two more aliases in October. That is a migration cadence, and it has a real engineering cost that the $5 price tag does not include. Add a two-year-old contract that permits only non-commercial use by default, a perpetual training licence over your inputs and outputs, and a $100 liability floor, and the cheapness starts to look like a price you pay somewhere else.

What would change this verdict is sound — and, separately, a contract as current as the documentation. When our audio benchmark runs, this page gains measured comparisons and the assessment table gains scores that mean something. Until then this is a verdict about a rate card, a deprecation calendar and a 2024 contract, worth exactly that much.

Frequently asked questions

Sources and what we could not verify

Every changing fact on this page was read from official Cartesia sources on September 1, 2026. The two headline pages — the deprecation table and the pricing page — were re-fetched and re-confirmed the same day. Where two official pages disagree, both readings are published above and neither is reconciled.

What we could not verify

Honesty about gaps beats a page that looks complete. Genuinely unresolved as of September 1, 2026, recorded rather than guessed:

• Whether the Voice Changer endpoints are actually off. Their published sunset date passed twelve days before our check, while the marketing site still publishes the capability page. Cartesia does not say which is current.
• What the Pro tier’s commercial licence actually covers. No page defines its scope, whether it survives cancellation, or how it treats work produced while subscribed.
• Which subscription tiers satisfy Terms §4.1. The contract never names one; only a pricing bullet does.
• Any requests-per-second or requests-per-minute rate limit. Only concurrency limits are published.
• Any service-level agreement or uptime commitment. None appears on the pages we read.
• Vendor latency figures in a form we would publish. Cartesia markets real-time performance without a figure we could cite as a specific documented claim.
• Whether the June 14, 2024 Terms have been superseded elsewhere. That is the date displayed on the page we read.
• The exact credits-to-characters conversion. Cartesia publishes credits and approximate minutes, not a character rate; our per-minute figures are computed from its own approximations.

Change log

September 1, 2026 — First publication. All prices, terms, sunset dates and feature statements read from official Cartesia sources on this date; the sunset and pricing pages re-confirmed the same day.

Published September 1, 2026 · The October 20, 2026 sunsets will be re-checked and recorded here after that date. Benchmark audio, measured costs and scores will be added with a dated entry when the audio benchmark runs (methodology). Independence note: this page contains no affiliate links, and no vendor paid for placement or influenced the order — see how we make money and our editorial policy.