Cartesia Review: The Fastest AI Voice Generator You Can Actually Build With

Verdict (TL;DR): Cartesia Sonic 3.5 is the developer's choice for low-latency text-to-speech. Sub-90ms response time, 42 languages, instant voice cloning from 10 seconds of audio. The free tier is genuinely useful. For Australians, the limitation is real: USD-only pricing, US data storage by default, no AUD invoicing, and a product built firmly for developers rather than non-technical users. If you're building a voice agent or podcast tool, Cartesia is worth serious consideration. If you want a plug-and-play voice tool with no code, look elsewhere.

Score: 13/25 — ⭐⭐⭐ Good

Last reviewed: 20 September 2026 by Adam.

What Cartesia is

Cartesia is a San Francisco AI company founded in 2023 by a team of Stanford AI PhD researchers — Karan Goel (CEO), Albert Gu, Arjun Desai, and Brandon Yang — who invented State Space Models (SSMs) during their research. The company's commercial products are built on that architecture.

The main product is Sonic, a text-to-speech (TTS) API. The current release is Sonic 3.5, launched in mid-2026. Alongside it sits Ink-2, a speech-to-text model, and Line, a voice agent platform. For this review, the focus is Sonic — Cartesia's flagship and the reason most people come looking.

The core claim is speed. Cartesia says Sonic 3.5 delivers sub-90ms time-to-first-audio — meaningfully faster than competing APIs, which typically land between 300ms and 800ms. That gap matters for real-time voice agents, interactive audiobooks, or anything where a noticeable pause breaks the experience.

Cartesia serves developers and businesses building voice-first products. It is not a self-contained no-code voice studio. You need to integrate via API.

What it costs in Australia

All pricing is in USD. There is no AUD billing. Australian users pay in USD and will be charged GST on top if their business is not registered for GST — check with your accountant. Exchange rate used below: 1 USD ≈ 1.40 AUD (estimate, September 2026).

Prices shown are monthly billing. Annual billing offers a discount of approximately 20%.

Voice agent (Line) calls are billed at USD $0.06 per minute on all plans, plus USD $0.014/min for telephony if you use a Cartesia-provided phone number. Speech-to-text (Ink-2) is included in each plan's credit allocation.

Pro voice cloning (higher fidelity, longer training audio) requires the Startup plan or above. Instant voice cloning — clone any voice from a 10-second clip — is available from the Pro plan ($5/mo).

Source: cartesia.ai/pricing, fetched 26 June 2026. Prices in USD; AUD conversions are estimates only. GST treatment depends on your ABN status — verify with your accountant.

How it works in practice

You sign up, get API credentials, and call the Sonic endpoint with your text. The API returns audio in your chosen format (MP3, WAV, PCM). You can stream the output — which is how the sub-90ms latency claim works in practice. First audio bytes arrive in under 90ms; the full clip takes longer depending on length.

Voice cloning works by uploading a short audio sample. Instant cloning needs 10 seconds. The Pro cloning tier (Startup and above) uses a longer sample and produces higher speaker similarity — better for a brand voice you'll use at scale.

The play.cartesia.ai playground lets you test voices without writing code. It's clean and functional. You can hear the difference between Sonic models and preview cloned voices before deploying. That's useful. What it is not is a full production tool — there's no built-in editor, no project management, no collaboration features aimed at content teams.

For voice agents, Line connects Sonic TTS with Ink STT and a large language model of your choice. You define a prompt and personality; the agent handles the real-time call loop. LLM usage is currently free for UI-created agents — though Cartesia notes this is "for a limited time."

In 42 languages, Sonic 3.5 handles English particularly well. Australian-accented English voices are available in the voice library, though the selection is smaller than the global voice pool. The quality is high — expressive, natural pacing, good handling of numbers and alphanumerics (phone numbers, order IDs, ABNs).

What it does well

Latency is genuinely class-leading. Sub-90ms time-to-first-audio is not marketing puff — independent benchmarks from Artificial Analysis and others confirm Cartesia leads the field here. For voice agents where a 1-second pause feels awkward, this matters.

The free tier is real. 27 minutes of TTS per month without a credit card. Enough to prototype properly before committing to a plan.

Voice cloning from a 10-second clip is fast and surprisingly clean. The output isn't flawless on short clips, but it's good enough for demos and internal tools.

42 languages at native quality. English, Spanish, French, German, Japanese, and 37 more — each trained rather than translated. That's useful for AU businesses serving multilingual customers.

SOC 2 Type II certified, GDPR compliant. Enterprise-grade security foundations are in place, which matters for regulated industries.

The Startup plan at $49/mo (≈ AUD $76/mo) is reasonable value if you're running a serious voice project. Pro voice cloning and 1,667 TTS minutes per month covers a lot of production use.

What it doesn't do well

It's an API, not an app. If you're a podcaster, marketer, or small business owner who wants to generate voiceovers without writing code, Cartesia is not for you. The playground is a preview tool, not a production studio. ElevenLabs or Murf better serve non-technical users.

The free plan has no commercial licence. You cannot use free-tier output in revenue-generating content. The Pro plan ($5/mo) unlocks commercial use — a low bar, but worth noting.

No AUD pricing. Every plan is in USD. Australian businesses pay the conversion plus potential GST. Invoices arrive in USD. For small businesses, this is a friction point.

Data residency is unclear for Australian users. Cartesia offers regional API endpoints and mentions in-region processing for enterprise customers, but standard plans do not guarantee Australian data storage. Your audio prompts and outputs route through US infrastructure by default. Businesses with data localisation obligations should clarify this before using the API. (verify with Cartesia sales)

Support below Scale tier is self-serve. Free, Pro, and Startup plans rely on documentation and a help centre. Priority support only unlocks at Scale ($299/mo USD). For Australian businesses that need a support SLA, that's a significant jump.

No built-in Australian business integrations. No native Xero, MYOB, Canva, or common AU workflow connectors. You build the integrations yourself.

Voice cloning consent is your responsibility. Cartesia's Acceptable Use Policy prohibits cloning voices without consent. In Australia, cloning a real person's voice without consent may breach the Privacy Act 1988, the Australian Consumer Law (misleading conduct), and emerging state-level personality rights. The buyer carries this risk, not Cartesia.

Australian fit

Cartesia is a US developer tool. It works anywhere with an internet connection and an API key, and the product quality is high. But the Australian-specific fit is limited:

• No AUD pricing — USD billing only
• No AU data residency guarantee on standard plans
• No GST-aware invoicing
• No AU-specific voice assets beyond a handful of accents in the general voice library
• No AU business integrations

That said, for Australian developers building voice agents — customer service bots, interactive IVR systems, real-time translation tools — Cartesia is a legitimate option. The latency advantage is real, the free tier lowers the entry cost, and the API is well-documented.

For voice-cloning use cases: any cloning of an identifiable Australian person's voice without their explicit written consent carries legal and reputational risk. Get consent in writing. Keep records. Disclosure when AI voice is used in consumer-facing content is increasingly expected and may become mandatory under AU consumer law reforms. This is your problem to solve, not Cartesia's.

Alternatives

If Cartesia doesn't fit your needs, consider:

• ElevenLabs — better non-technical UI, larger voice library, stronger emotional range. More expensive at scale. The go-to for content creators.
• Murf — a full voice studio with script editor, timeline, and collaboration tools. Better for marketers and L&D teams who don't want to code.
• Resemble AI — another API-first voice platform with strong cloning and good enterprise features. Note that its published pricing now covers deepfake detection only; the voice models are MIT open source under the Chatterbox name, and a commercial voice contract goes through sales.

Bottom line

Cartesia Sonic 3.5 is technically impressive. The speed is real, the voice quality is high, and the free tier gives you room to test before spending. For Australian developers building voice applications, it belongs on the shortlist.

The problem is everything around the core product. USD pricing, no AU data residency guarantee on standard plans, no business-user-friendly interface, and thin local support make Cartesia a specialist tool rather than a general recommendation. It earns a solid score for feature depth and value at the developer tier, but Australian fit pulls the overall score down.

If you're a developer building real-time voice features into a product, try the free tier today. If you're a non-technical business owner who wants to generate voice content, start with ElevenLabs or Murf instead.

Score

Frequently asked questions

Is Cartesia available in Australia?

Yes. Cartesia's API is accessible from Australia. Pricing is in USD and data is processed through US infrastructure by default on standard plans. Enterprise customers can discuss in-region processing. There is no AUD pricing or GST-aware invoicing.

How fast is Cartesia Sonic 3.5 compared to ElevenLabs?

Cartesia Sonic 3.5 achieves sub-90ms time-to-first-audio — significantly faster than ElevenLabs, which typically lands between 300ms and 500ms. That gap matters for real-time voice agents where response delay is noticeable. For pre-recorded voiceovers, the latency difference is largely irrelevant.

Can I use Cartesia voice cloning for commercial content in Australia?

Commercial use requires the Pro plan ($5/mo USD) or higher — the free tier does not include a commercial licence. Voice cloning of real people without their explicit consent is prohibited by Cartesia's Acceptable Use Policy and may also breach Australian privacy law and the Australian Consumer Law. Always obtain written consent before cloning an identifiable person's voice.

Sources

• Cartesia pricing page: cartesia.ai/pricing — fetched 26 June 2026 — Wayback 30 June 2026
• Cartesia company page: cartesia.ai/company — fetched 26 June 2026 — Wayback 6 June 2026
• Cartesia Sonic 3.5 product page: cartesia.ai/sonic — Wayback 4 September 2026
• Cartesia launch blog (Sonic-3.5 and Ink-2): cartesia.ai/launch — Wayback 3 September 2026
• Cartesia GDPR compliance blog: cartesia.ai/blog/gdpr-compliance — Wayback 14 June 2026
• Cartesia Acceptable Use Policy: cartesia.ai/legal/acceptable-use — Wayback 14 June 2026
• Artificial Analysis TTS benchmarks: artificialanalysis.ai/text-to-speech — Wayback 8 February 2026
• AUD/USD conversion: approximately 1 USD = 1.40 AUD (estimate, September 2026). GST not included in converted figures.

The AI Directory Australia is an independent editorial publication. We don't accept payment for positive reviews. Read our Methodology and Affiliate Disclosure.