Cartesia vs Rime (2026): Fastest Voice API or Most Human Voices?

Pricing figures: Cartesia's USD prices are converted at 1 USD ≈ 1.40 AUD (verified 5 September 2026), as in our Cartesia review; Rime's at US$1 ≈ A$1.43 (RBA, 7 October 2026), as in our 8 October 2026 Rime review. All are shown ex-GST. A GST-registered Australian business that supplies its ABN is not charged GST on an imported digital service; other buyers should allow ~10% on top (Methodology).

The short version: This was speed against warmth, and the gap has narrowed. Cartesia sells fast text-to-speech — sub-90ms time-to-first-audio, independently benchmarked — plus speech-to-text and a voice-agent layer on one bill. Rime sells human-sounding voices built for spoken business interactions: phone bots, IVR, outbound calling. Its new flagship, Coda, quotes a 96 ms median time to first audio and its Mist v3 model 37 ms, so Cartesia no longer clearly owns the stopwatch. Both are developer-only APIs. Both bill in USD with no AUD invoicing, and neither offers Australian data residency in the cloud. But Rime now ships six Australian voices. For the typical Australian buyer — a developer or product team wiring up a realtime voice agent — Rime now takes it, 15/25 to Cartesia's 13/25. Pick Cartesia if you want the whole loop from one vendor.

Winner changed, 8 October 2026. Our June edition named Cartesia the winner on latency, against Rime's Arcana model at ~200ms in the cloud. Arcana was retired on 15 August 2026, and Rime re-scored 14/25 → 15/25 on its Australian voices. The reasons are set out in the verdict.

We have reviewed both tools in full. This piece does not re-score them. It puts them side by side for an Australian developer choosing one, and names who should buy which.

• Cartesia review — 13/25, ★★★ Good
• Rime review — 15/25, ★★★ Good

Earlier editions of this comparison: 28 Jun 2026. This is edition 2, dated 8 Oct 2026. We keep every edition so you can see how the tool and our verdict have moved — how and why.

The tension: the stopwatch versus the ear

Cartesia and Rime are both San Francisco voice companies, both founded by research teams, both selling TTS to developers. But they optimise for opposite things.

Cartesia was built by Stanford AI researchers who invented State Space Models. Its flagship, Sonic 3.5, is engineered around one number: time-to-first-audio under 90ms. That is roughly three to eight times faster than most rival APIs, which land between 300ms and 800ms. In a realtime voice agent, that gap is the difference between a reply that lands and a pause the caller fills with "hello?".

Rime was built by linguists. The founding team recorded the full mess of spontaneous human speech — breaths, filler words, laughter, disfluencies — and trained models to reproduce it. Its previous flagship, Arcana v3, was retired from Rime's cloud on 15 August 2026. Requests for it now run on Coda, launched on 19 May 2026 and trained on full-duplex conversations between real people. Rime's pitch is unchanged: the voice on the other end of a sales or service call should sound like a person, not a machine.

So the choice is genuinely about temperament. Do you want the most complete, fastest-benchmarked stack, or the voice that sounds most like a person? For an Australian buyer, that question gets sharpened by two things neither vendor solves out of the box: what it costs in dollars, and where the audio lives.

What they cost in Australia (AUD, ex-GST)

Both price in USD. Neither bills in AUD or issues GST-aware invoices. Cartesia's conversions use ~1.40 (5 September 2026); Rime's use US$1 ≈ A$1.43 (RBA, 7 October 2026). GST treatment depends on your ABN and registration status — verify with your accountant.

Cartesia sells subscription tiers with bundled minutes. Rime sells usage-based, per-character rates with free minutes to start. They do not line up cleanly, which is itself a planning point: Cartesia is easier to budget at a fixed monthly spend; Rime scales linearly with volume.

Cartesia pricing: cartesia.ai/pricing, fetched 30 June 2026 and re-checked by browser 1 October 2026 — Wayback snapshot 30 June 2026. Rime pricing: rime.ai/pricing, read 8 October 2026 per our Rime review — Wayback snapshot 30 September 2026. AUD figures are conversions, ex-GST.

Rime has simplified to two plans, Starter and Enterprise, with per-model rates published openly. The earlier "Growth" plan, its launch promo and the Arcana rate are all gone. Rime's pricing page states its free allowance two ways: "~800 minutes free" on the plan card and 3,000 in its FAQ. Plan on the lower figure until your dashboard shows otherwise. Cartesia's free tier allows 8 concurrent TTS requests. Confirm current rates before you budget; both vendors move quickly.

On headline cost, neither is expensive at small scale. Cartesia's $5 Pro tier is the cheapest serious subscription in the category. Rime's free Starter minutes are a generous evaluation window. The cost question only bites at volume, and there it depends entirely on your minutes — model it against your own traffic, not a headline.

Feature diff: where they actually differ

The honest read: these are not the same product wearing different badges.

Cartesia is the more complete stack for building a realtime agent yourself. It bundles TTS, speech-to-text, and an agent layer (Line) that connects them to your chosen LLM. You can stand up a working voice loop without stitching three vendors together. Its latency is independently benchmarked. Rime's own figures now match or beat it, but they are Rime's numbers, not an independent test, and we have not timed either from Australia.

Rime is narrower and deeper. It does TTS only, and the sound is the point. Rime says Coda tops human evaluations for voice quality; that is the vendor's claim, and we have not yet run our own listening test on Coda. If your differentiator is a caller hanging up thinking they spoke to a person — or hearing a local accent — Rime's voices earn their place. But you supply the rest of the stack — speech-to-text, orchestration, telephony — yourself, typically via LiveKit, Pipecat, Vapi or Twilio.

So the feature gap maps neatly to the brief. Want fewer moving parts and one bill? Cartesia. Want the most human voice, or an Australian one, for phone and contact-centre work, and already have an orchestration layer? Rime.

Australian fit: Rime pulls a point ahead

Neither tool does well on the criterion The AI Directory weights most heavily, but they no longer tie. Rime now scores 3/5 on Australian Fit, Cartesia 2/5. Neither reaches the 4/5 that the Australian Fit gate requires; neither could be named a category Top Pick on local fit as it stands.

Data residency. Both process audio through US infrastructure by default. Cartesia mentions in-region processing for enterprise customers but guarantees nothing on standard plans; your prompts and outputs route through the US. Rime's cloud runs in two regions, US West and US East, with one advantage: it publishes on-prem and VPC deployment for Enterprise customers, which keeps audio inside your own infrastructure entirely. That path resolves the residency problem — but only at enterprise cost and complexity, and only if you can build and run it.

Australian voices. This is what moved. Cartesia has a handful of Australian-accented English voices in its general library, though the local selection is thin. Rime's Coda catalogue now carries six named Australian voices — two male (marlu, drownlight) and four female (eucalyptus, perth, southvale, matilda). Its faster Mist v3 model has none. For an AU contact centre where callers respond better to a local-sounding voice, Rime is now the stronger choice.

Pricing and GST. Identical story. USD billing, no AUD invoicing. A GST-registered Australian business that supplies its ABN is not charged GST on these imported services; other buyers add about 10%. Treat the USD-to-AUD conversion as a moving cost.

Support. Both are self-serve below their top tiers, with no Australian-timezone presence. Cartesia unlocks priority support only at Scale (USD $299/mo). Rime's self-serve users get documentation and a public Slack; meaningful support sits behind Enterprise. Whichever you pick, expect to live in the documentation.

Daylight also opens on reliability and support, where Rime leads (3/5 to Cartesia's 2/5) on its HIPAA BAA and published on-prem path — useful if you are a regulated AU buyer who can fund the deployment. One caution cuts the other way: Rime's line-up moves fast. Arcana went from flagship in our June review to retired from the cloud by mid-August. If you build on a Rime model, plan for migrations.

Honest limitations

Neither tool is for a non-technical buyer. If you are a podcaster, marketer, or small-business owner who wants to type a script and download a voiceover, both are the wrong shelf — start with ElevenLabs or Murf, which we review separately. There is no studio, no timeline editor, no no-code path to a finished file here.

And the realtime-agent build is never just the voice. Cartesia gets you closest to a complete loop, but you still write the integration. Rime gets you the more human voice, but you assemble the rest. Budget engineering time, not just API credits.

Who buys which

Buy Rime if you are building a phone or contact-centre voice agent where sounding human — or Australian — matters, and you already run an orchestration layer (LiveKit, Pipecat, Vapi, Twilio). Coda's six Australian voices and its quoted latency make it the typical Australian builder's pick now. Choose Rime especially if you are a regulated AU business that can fund on-prem or VPC deployment, because that is the cleaner answer to the data-residency question than anything Cartesia offers on a standard plan.

Buy Cartesia if you want the whole loop from one vendor. Bundled speech-to-text and the Line agent layer mean you can prototype a realtime agent quickly without stitching three services together, and its sub-90ms latency is backed by independent benchmarks rather than the vendor's word. The $5 Pro tier and a real free tier make it cheap to test, and a fixed subscription is easier to budget than per-character billing.

The verdict

For the typical Australian buyer in this category — a developer wiring up a realtime voice agent — Rime now wins. That is a change. Our June edition gave it to Cartesia on latency: sub-90ms against Arcana's ~200ms in the cloud. Arcana has since been retired, and Rime's replacement, Coda, quotes a 96 ms median time to first audio, with Mist v3 at 37 ms. Those are Rime's figures, not an independent benchmark, and from Australia every request to either vendor crosses the Pacific. But latency no longer clearly separates these two tools, and the rest of the ledger favours Rime: 15/25 to 13/25, Australian Fit 3/5 to 2/5 on the strength of six Australian voices, and a stronger compliance and on-prem story.

Cartesia remains the right call for a narrower buyer: the team that wants TTS, speech-to-text and an agent layer on one bill, at a fixed monthly spend. Time both from an Australian host before you commit.

The verdict comes with the same asterisk attached to both reviews: neither tool is excellent for Australians yet. Rime scores 3/5 on Australian Fit and Cartesia 2/5; both bill in USD, and neither guarantees Australian data residency without an enterprise contract. Neither clears the Australian Fit gate, so neither can be a category Top Pick. Pick on capability, then read the contract closely on data and price before you sign.

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

Frequently asked questions

Cartesia vs Rime: which is faster for a realtime voice agent?

It is close now, and the numbers are not like-for-like. Cartesia's Sonic 3.5 delivers sub-90ms time-to-first-audio, backed by independent benchmarks. Rime quotes a 96 ms median for its Coda model and 37 ms for Mist v3; those are Rime's own figures. Arcana, the Rime model we compared in June, was retired on 15 August 2026. Both run in US regions, so time requests from your own Australian host before you commit.

Cartesia vs Rime: which sounds more human?

Rime, on its design brief. Its Coda model is trained on full-duplex conversations between real people, by a company founded by linguists, and it now includes six Australian voices. Rime says Coda tops human evaluations; we have not yet run our own listening test on it. Cartesia's voices are high quality, but its core advantages are speed and a bundled stack, not naturalness.

Do Cartesia or Rime offer Australian data residency?

Neither guarantees Australian data residency on standard plans. Both process audio through US infrastructure by default; Rime's cloud runs in US West and US East only. Rime publishes on-premise and VPC deployment for Enterprise customers, which can keep audio inside your own infrastructure. Cartesia mentions in-region processing for enterprise customers but guarantees nothing on standard plans. In our reviews Rime scores 3 out of 5 on Australian Fit and Cartesia 2 out of 5.

How much do Cartesia and Rime cost in Australian dollars?

Both price in USD with no AUD invoicing. Cartesia runs on subscription tiers: Free, Pro at USD $5/mo (about AUD $7.00), Startup at USD $49/mo (about AUD $69), and Scale at USD $299/mo (about AUD $419), plus USD $0.06/min for its Line voice agent. Rime is usage-based: Mist v3 at US$0.03 per 1,000 characters (about A$0.043) and Coda at US$0.05 (about A$0.072), with free minutes on its Starter plan. A GST-registered Australian business that supplies its ABN is not charged GST on these imported services; other buyers add about 10%. Confirm current rates on each vendor's pricing page.