# Cartesia review (2026): What it is and why enterprises choose Bland Speech

By Samuel Oram-Jones

Cartesia sells text to speech in credits, one credit per character, and says a minute of Sonic audio uses about 750 to 800 of them. That conversion is the first piece of arithmetic in any Cartesia purchase, because every plan, every overage rate, and every cloning slot is quoted in credits rather than dollars per character.

Sonic is a fast model with a long language list, and a team that wants either of those has a real reason to buy it. A team whose customers hear the voice on a phone line is weighing a different question, which is how human the voice sounds, and on that one Design Arena's listening test of 4 August 2026 ranked Cartesia's Sonic 3.5 sixth of 13 AI voices and Bland Speech v3 first.

## What is Cartesia, and how does it work?

Cartesia is a company that sells Sonic, a text to speech model, next to speech to text and a voice agent product called Managed Agents, and as of November 2026 its current model is Sonic 3.6. Cartesia released Sonic 3.6 on 27 August 2026 and says it speaks 44 languages from one model. Sonic 3.5 and Sonic 3 are still served, and Cartesia says Sonic 3.6 is fully backwards compatible with Sonic 3.5.

A developer reaches Sonic through three endpoints: POST /tts/bytes, POST /tts/sse, and a WebSocket at /tts/websocket. All three stream audio as it is produced, but Cartesia says only the WebSocket accepts text in pieces, which is what a voice agent needs when a language model writes a reply a few words at a time. Cartesia says Sonic delivers first audio in under 90ms, and Bland publishes 199ms to first audio, so each company's figure is its own claim rather than a race anyone ran.

### What features does Cartesia add?

Cartesia gives a developer speed from 0.6 to 1.5, volume from 0.5 to 2.0, and emotion controls, and the emotion controls are in beta and work in English only. Pronunciation dictionaries let a team set how a product name or a medication is said, then pass the dictionary by id on any call. Word or segment timestamps come back on the WebSocket and SSE endpoints, and not on the bytes endpoint, which is useful for captions or for knowing where a caller interrupted.

Voice cloning has two levels. An instant clone needs 10 seconds of audio and a Pro plan or above, with up to 60 seconds helping it keep an accent. A Pro Voice Clone needs at least 30 minutes of audio, trains for up to 3 hours, and starts on the Startup plan, with 2 slots on Startup and 4 on Scale. Training is free and speech from a clone costs 1 credit per character, and Cartesia adds that speed and volume controls don't apply to a Pro Voice Clone, because it learns pacing and loudness from its dataset.

### How easy is Cartesia to use?

Cartesia is easy to start with and takes more planning to run in production. Cartesia publishes a side-by-side comparison of its three endpoints, so a developer can pick one in a few minutes. The planning comes from the plan system, because commercial use, instant cloning, professional cloning, and the overage rate each depend on which plan a team holds.

Cartesia also retires models on dated schedules. The original Sonic models ended on 1 June 2026, and sonic-2, sonic-turbo, and a Sonic 3 snapshot carried a sunset date of 20 October 2026. Cartesia stops hosting code-first Line SDK agents on 1 January 2027, which sends those teams to Managed Agents. None of that is unusual for a vendor moving fast, and all of it lands on the team that built on an older snapshot.

## How much does Cartesia cost?

Cartesia costs $5 a month for 100,000 credits on Pro, $49 for 1.25 million on Startup, and $299 for 8 million on Scale, with Enterprise custom, and one credit buys one character of Sonic speech. The Free plan gives 20,000 credits a month, about 27 minutes of Sonic 3.6 audio, which Cartesia describes as for evaluation and personal use, and the commercial use license starts on Pro.

Past a plan's allowance, overage is opt-in and costs $65, $45, or $38 per million credits on Pro, Startup, and Scale, billed when the plan renews. With overages off, requests stop at zero credits. Unused credits roll over up to twice the monthly amount. A few small items also draw credits: each break tag counts as 1 credit, and adding an accent to a voice costs 225 credits.

Cartesia bills voice agents separately, at $0.03 a minute on every self-serve plan, with telephony on a Cartesia number at $0.014 a minute and language model usage on top.

### What does a month of speech cost on Bland Speech and on Cartesia?

Bland Speech costs $0.04 per 1,000 characters, which is $40 per million, with no monthly plan, and a new account gets 50,000 characters free once. Credits never expire. Enterprise pricing is custom, set with Bland's sales team.

At small and middle volumes Bland Speech's standard rate is lower, and at 8 million characters a month Cartesia's Scale plan costs $21 less, as long as a team uses the full allowance every month.

| Characters a month | Bland Speech | Cartesia |
| --- | --- | --- |
| 100,000 (about 2 hours of audio) | $4 | $5 on Pro |
| 1 million (about 20 hours) | $40 | $49 on Startup |
| 8 million (about 175 hours) | $320 | $299 on Scale |

<details><summary>What these rows mean</summary>

- **Characters a month:** The hours are rough, using Cartesia's own figure of 750 to 800 characters for a minute of Sonic speech.
- **Bland Speech:** $0.04 per 1,000 characters, so 100,000 characters is $4 and 8 million is $320, with no plan fee.
- **Cartesia:** The cheapest self-serve plan whose credits cover the volume, at 1 credit per character. Telephony, language model usage, and taxes are left out of both columns.

</details>

## Is Cartesia secure enough for enterprise calls?

Cartesia says it is SOC 2 Type II certified, and its trust center names PCI DSS 4.0.1, HIPAA, SOC 2 Type 2, and GDPR. Cartesia words HIPAA as eligible with BAAs available for healthcare customers, and it puts DPAs, BAAs, SSO, security questionnaires, and custom concurrency on Enterprise. Cartesia also says zero data retention is available for enterprise customers across eligible services.

On deployment, Cartesia offers Sonic on-premise inside a customer's data center, including air-gapped environments, in a customer's own VPC on AWS, Google Cloud, or Azure, or through OEM licensing, all under enterprise contracts. Dedicated regional deployments are available for the United States, European Union, United Kingdom, India, and Australia.

Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted on 17 November 2026, and its [trust page](/trust-security) says speech synthesis runs inside the FedRAMP boundary. Bland runs its own model and speech, so no outside AI company sits in a call. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland.

## Which sounds more human, Cartesia or Bland Speech?

Bland Speech v3 scored 1365 in the blind tests Design Arena published on 4 August 2026, 269 points above Cartesia's Sonic 3.5 at 1096. In those [Audio Realism Bench results of 4 August 2026](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/), Bland Speech v3 ranked first of 13 AI voices, Sonic 3.5 ranked sixth, and a human recording scored 1500.

Listeners in that test hear two clips of the same line from two models, without being told which is which, and pick the one that sounds more like a real person. The test's prompts include a Phone Agents category for customer support and routing calls, which is the work a text to speech model does on a phone line. Cartesia released Sonic 3.6 on 27 August 2026, after those results, so Sonic 3.6 wasn't scored in them. Design Arena's article says "Second only to human speech is Bland Speech v3 by Bland AI, with an Elo of 1365."

## Is Cartesia worth it, or should a team choose Bland Speech?

Cartesia is worth it for a team that needs 44 languages, word timestamps, or an on-premise deployment, and Bland Speech is the better choice for a phone line where the voice is a caller's first impression. A team comparing Cartesia with other vendors can read [Cartesia alternatives](/blog/cartesia-alternatives), and a team already on Cartesia can follow [how to migrate from Cartesia](/blog/migrate-from-cartesia).

### Cartesia pros and cons

<div class="cmp-proscons" data-rival="cartesia"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>44 languages from one model, Sonic 3.6</li><li>First audio in under 90ms, as Cartesia states it</li><li>Word or segment timestamps on its WebSocket and SSE endpoints</li><li>Pronunciation dictionaries passed by id on any call</li><li>Mu-law and A-law at 8 kHz for phone lines</li><li>On-premise, air-gapped, and VPC deployment under enterprise contracts</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Sonic 3.5 ranked 6th of 13 AI voices on Audio Realism Bench on 4 August 2026</li><li>Every price is in credits, which a team converts to characters and minutes before forecasting</li><li>Instant clones start on the $5 Pro plan, and professional clones start on the $49 Startup plan</li><li>Emotion controls are in beta and English only</li></ul></div></div>

### Why do enterprises choose Bland Speech over Cartesia?

Enterprises choose Bland Speech when callers will judge a voice in the first second and when finance wants one standard rate instead of a plan ladder. Bland Speech v3 ranked first on Design Arena's listening test of 4 August 2026, and its price is $0.04 per 1,000 characters with no allowance to manage.

On the developer side, a team already using OpenAI's audio speech API can point it at POST /v2/audio/speech by changing a base URL, a key, and a model name, and a team streaming a language model's output uses WSS /v2/tts/ws, where a new context interrupts the turn that is playing. Phone audio comes as mu-law at 8 kHz, or as PCM from 8 to 48 kHz. Bland Speech has a first-party Pipecat service and a LiveKit Agents plugin for Python, and the [text to speech API](/text-to-speech-api) page has the code.

Performance tags in square brackets, such as [laughs], [sighs], or [say warmly], tell the model how to deliver the words after them. An instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio after Bland's consent and rights review, and every clone is private to the customer's organization. [Bland Speech](/speech) has a studio for trying all of it.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Bland Speech v3 ranked first of 13 AI voices on Audio Realism Bench on 4 August 2026</li><li>One standard rate, $0.04 per 1,000 characters, with no monthly plan</li><li>An OpenAI-compatible endpoint, so an existing integration changes a base URL, a key, and a model name</li><li>Performance tags for how a line is read</li><li>Mu-law at 8 kHz and PCM from 8 to 48 kHz for phone lines</li><li>A first-party Pipecat service and a LiveKit Agents plugin for Python</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>More than a hobby project needs, because Bland Speech is built for production calls</li></ul></div></div>

### Verdict: Cartesia or Bland Speech for enterprise calls?

For a business that puts synthetic speech on a phone line, Bland Speech is the pick, because it ranked first on voice realism in Design Arena's 4 August 2026 results and bills one standard rate. Cartesia is the pick for a team that needs 44 languages from one model, word timestamps, or a model running in its own data center, and Cartesia's under-90ms claim matters to a team that counts every millisecond. A team pricing both can start with [text to speech API pricing in 2026](/blog/text-to-speech-api-pricing), and the [Bland Speech and Cartesia voice cloning comparison](/compare/bland-speech-vs-cartesia-voice-cloning) goes deeper on clones. More reviews sit in the [text to speech reviews](/compare/text-to-speech/reviews) collection, and the [text to speech comparison hub](/compare/text-to-speech) covers every vendor.

## Frequently asked questions

### Is Cartesia good for phone calls?

Cartesia works for phone calls because it returns mu-law and A-law at 8 kHz and streams first audio quickly, with Cartesia claiming under 90ms. The catch is voice realism: Sonic 3.5 ranked 6th of 13 AI voices in Design Arena's 4 August 2026 results, against first for Bland Speech v3, and Sonic 3.6 was released after them.

### How much does Cartesia cost per month?

Cartesia costs $5 a month on Pro, $49 on Startup, and $299 on Scale, plus custom Enterprise pricing and a Free plan with 20,000 credits. One credit buys one character of Sonic speech, so Startup's 1.25 million credits cover 1.25 million characters.

### How many credits is a minute of Cartesia speech?

A minute of Sonic audio uses about 750 to 800 credits, by Cartesia's own figure. Free's 20,000 monthly credits therefore cover about 27 minutes, and Scale's 8 million cover about 10,667.

### Does Cartesia clone voices?

Yes. An instant clone needs 10 seconds of audio and a Pro plan or above, and a Pro Voice Clone needs at least 30 minutes of audio and a Startup plan or above. Bland Speech clones from about 10 seconds for an instant clone and 30 minutes or more for a professional one, with no monthly plan and a consent and rights review for professional voices.

### Is Cartesia HIPAA compliant?

Cartesia names HIPAA among its certifications and words it as HIPAA-eligible, with BAAs available for healthcare customers. Cartesia puts BAAs on its Enterprise plan, so a healthcare team needs that tier. Bland is HIPAA compliant with a signed BAA.

### Is Cartesia faster than Bland Speech?

By each company's published figure, yes. Cartesia says Sonic delivers first audio in under 90ms, and Bland publishes 199ms to first audio. Both stream over a WebSocket, and neither figure comes from a test a third party ran.

### What is the best Cartesia alternative?

Bland Speech is the best Cartesia alternative for phone calls and voice agents, because it ranked first of 13 AI voices on Audio Realism Bench on 4 August 2026 and bills $0.04 per 1,000 characters. Teams wanting more choices can read the [best Cartesia alternatives](/blog/cartesia-alternatives).

### Does Cartesia have a free plan?

Yes. Cartesia's Free plan includes 20,000 credits a month, about 27 minutes of Sonic 3.6 audio, with no overages, and Cartesia describes it as for evaluation and personal use. Bland Speech gives a new account 50,000 characters free once, and the same rate applies after that.

## Sources

Cartesia's pricing page, Sonic page, documentation on models, endpoints, output formats, cloning, and regional deployments, and trust center. Bland's [pricing](/pricing) and trust and security pages and Bland Speech documentation. Design Arena's Audio Realism Bench article of 4 August 2026.
