# Cartesia vs Rime (2026): Which latency number can a phone agent plan around?

By Samuel Oram-Jones

Rime says the median time to first audio for Coda v2 is 30ms, measured at the server with no network in the way, and Rime also publishes a benchmark in which Coda's median is 96ms for a single request and 150ms with 12 requests running at once. Neither number is wrong, because they measure different things under different load. A buyer reading them next to Cartesia's claim of under 90ms for Sonic needs the conditions as much as a figure.

As of November 2026, the choice between Cartesia and Rime also turns on how each bills, how many languages each speaks, and how a team gets a custom voice. Cartesia's Sonic 3.5 ranked 6th of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, Rime's voices weren't in those results, and Bland Speech v3 ranked first at 1365, which is why [Bland Speech](/speech) belongs in the comparison.

## Which is better, Cartesia or Rime?

Cartesia is the better fit for a team that needs many languages or a custom voice made from a short clip, and Rime is the better fit for an English-led regulated team that wants billing per minute, a HIPAA compliant vendor that signs a BAA, and a dedicated VPC or on-prem option. Cartesia speaks 44 languages on Sonic 3.6 and clones an instant voice from 10 seconds of audio, while Rime serves 10 languages on Coda and sells custom voices as an enterprise order.

| Feature | Bland Speech | Cartesia | Rime |
|---|---|---|---|
| Audio Realism Bench, 4 August 2026 | 1st at 1365 (Bland Speech v3) | 6th at 1096 (Sonic 3.5) | Not in those results |
| Billing unit | Per 1,000 characters | Credits, one per character | Per minute of audio, about 1,000 characters a minute |
| Price | $40 per million characters, pay as you go | $5 to $299 a month, or Enterprise | $0.050 a minute on Flex for Coda v2, $0.040 on Scale at $100 a month |
| First audio, as each company publishes it | 199ms | Under 90ms | 30ms or 96ms for Coda, depending on the measure |
| Custom voice | Instant from about 10 seconds, professional from 30 minutes or more | Instant from 10 seconds, professional on Startup and above | Enterprise voice cloning ordered from Rime, 30 to 60 minutes of recordings |
| Phone audio, mu-law at 8 kHz | Yes | Yes | Yes |
| Compliance | SOC 2 Type II, HIPAA compliant with a signed BAA | SOC 2 Type II, BAAs on Enterprise | SOC 2 Type 2, HIPAA compliant, BAA available |

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

- **Audio Realism Bench:** Design Arena's blind listening test, in which listeners hear two clips of one line and pick the one that sounds more like a real person. A higher Elo rating means a voice won more of those votes.
- **Billing unit:** Cartesia counts one credit per character and draws credits from a monthly plan. Rime bills minutes of generated audio and says one minute is about 1,000 characters, so Rime's $0.050 a minute is close to $50 per million characters on Rime's own conversion.
- **First audio:** The time from sending text to hearing the first sound. Rime's 30ms is a median at the server without network time, and its 96ms is the median in Rime's benchmark for one request at a time. Bland publishes 199ms, and Cartesia says under 90ms.
- **Custom voice:** A voice built from a person's recordings. A BAA is a HIPAA business associate agreement, the contract a healthcare buyer needs before sending patient information to a vendor.

</details>

## Where does Cartesia fall short next to Rime?

Cartesia falls short for a team that wants usage to map straight onto a bill, because Cartesia's text to speech draws one credit per character from a monthly plan, and a break tag in a script costs a credit as well. Rime's Free plan includes $50 of usage a month, and Rime's Flex plan has no monthly fee, so a pilot starts without a subscription. Cartesia's Free plan is 20,000 credits a month, about 27 minutes of Sonic-3.6 speech, and Cartesia calls Free "for evaluation and personal use".

Cartesia also says "With overages off, new requests are blocked once your credits run out." Older models are on a clock too: sonic-2, sonic-turbo, and a Sonic 3 snapshot carry a 20 October 2026 sunset date, and Cartesia stops hosting Line SDK agents on 1 January 2027.

Cartesia's strengths are breadth and a quick start. Sonic 3.6 speaks 44 languages, Cartesia says first audio arrives in under 90ms, and a voice clone starts from 10 seconds of audio. Cartesia also offers on-prem and VPC deployment under enterprise contracts.

<div class="cmp-proscons" data-rival="cartesia"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>44 languages on Sonic 3.6</li><li>Instant cloning from 10 seconds of audio</li><li>Says first audio arrives in under 90ms</li><li>Scored in Design Arena's 4 August 2026 results, at 1096</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Credits drawn from a monthly plan, with requests blocked at zero when overages are off</li><li>Each break tag in a script costs a credit</li><li>Older Sonic models and Line SDK hosting have end dates</li></ul></div></div>

Cartesia is a poor fit for:

- A team that wants billing per minute of audio with no credits to track
- A script full of timed pauses, since each break tag is billed

## Where does Rime fall short next to Cartesia?

Rime falls short for a team that serves callers in many languages, because Coda's cloud API covers 10 languages (English, Arabic, Dutch, French, German, Hindi, Italian, Japanese, Portuguese, and Spanish), each voice serves one of them, and 185 of Coda's 313 voices are English. Mist v3, Rime's low-latency model, serves English, French, German, and Spanish.

Custom voices take a longer route. Rime sells enterprise voice cloning, ordered from Rime for Coda or Mist, from 30 to 60 minutes of clean recordings for a basic clone and two to five hours or more for stronger results, which is a project with a sales contact where Cartesia's is a plan upgrade.

Rime's latency story takes more reading than Cartesia's one line. Rime says its cloud API delivers "sub-200ms end-to-end latency under typical conditions", says the cloud adds about 25 to 50ms of network time to a model's own latency, and publishes a Coda median that grows from 96ms to 150ms when 12 requests run at once, while Mist v3 holds at 37ms. A team sizing a phone turn has to pick which of those numbers matches its load.

Rime's strengths sit with regulated buyers. Rime says it has been SOC 2 Type 2 compliant since May 2025 and HIPAA compliant since February 2024, with its most recent audit completed in March 2026, and signs MSAs and BAAs. Enterprise customers can run Rime in a dedicated VPC or fully on-prem, on an NVIDIA host with Docker Compose. Rime also streams over HTTP, WebSocket, and gRPC, sends word-level timestamps on WebSocket and gRPC for Coda, and documents LiveKit, Pipecat, Daily, and Vapi.

### What Rime costs

Rime's Free plan is $0 with $50 of usage every month across Prism, Mist v3, Coda v2, and speech to text. Flex has no monthly fee and bills Mist v3 at $0.030 a minute and Coda v2 at $0.050. Scale is $100 a month and takes 20% off, which makes Mist v3 $0.024 and Coda v2 $0.040 a minute, with the $100 credit rolling over up to $200. Enterprise is custom volume pricing billed annually. Rime retired Arcana from its cloud API on 15 August 2026, and Mist v1 and v2 retire on 3 December 2026.

<div class="cmp-proscons" data-rival="rime"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>HIPAA compliant, SOC 2 Type 2, and BAAs available</li><li>Dedicated VPC or on-prem for enterprise customers</li><li>$50 of free usage a month, then billing per minute</li><li>HTTP, WebSocket, and gRPC streaming with word-level timestamps on Coda</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Two different first-audio figures for Coda on Rime's own pages</li><li>10 languages on Coda, and 185 of its 313 voices are English</li><li>Custom voices ordered from Rime, from 30 to 60 minutes of recordings</li><li>Mist v1 and v2 retire on 3 December 2026</li></ul></div></div>

Rime is a poor fit for:

- A contact center that needs more than 10 languages
- A team without 30 minutes of clean recordings for a custom voice

## Which sounds more human, Cartesia or Rime?

Only Cartesia can be placed, because Rime's voices weren't in Design Arena's Audio Realism Bench results of 4 August 2026. In those results Cartesia's Sonic 3.5 ranked 6th of 13 AI voices at 1096. Rime released Coda in May 2026 and Mist v3 in March 2026, both before those results, and neither was scored.

Cartesia released Sonic 3.6 on 27 August 2026, after the results, so the scored model is older than the one Cartesia sells now. Bland Speech v3 ranked first in those results at 1365, human recordings scored 1500, and Design Arena explains [how the listening test works](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/). With no shared score for Cartesia and Rime, a team picking between them has to listen to both on its own scripts, in the voices it plans to ship.

## Why do people comparing Cartesia and Rime also weigh Bland Speech?

People comparing Cartesia and Rime also weigh Bland Speech because Bland Speech v3 ranked first in the 4 August 2026 listening results, where Sonic 3.5 placed 6th and Rime wasn't tested, and Bland Speech bills one standard rate. Bland Speech bills $0.04 per 1,000 characters ($40 per million), pay as you go with no monthly plan, and 50,000 free characters once for a new account. That rate sits close to Rime's Coda v2 at $0.040 a minute on Scale and inside Cartesia's $37.38 to $50 plan range.

Bland publishes 199ms to first audio, slower than both rivals' claims, as each company publishes it. Bland Speech outputs mu-law at 8 kHz or PCM from 8 to 48 kHz, streams over HTTP and a WebSocket where a new context interrupts the turn that is playing, and has a first-party Pipecat service and a LiveKit Agents plugin for Python. An instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, and every clone is private to the customer's organization.

Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA, owns its speech model, and holds FedRAMP 20x Class C certification at Moderate, granted 17 November 2026, with speech synthesis inside its FedRAMP boundary. Enterprise pricing is custom, set with Bland's sales team.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Ranked first in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365</li><li>One standard rate, $0.04 per 1,000 characters</li><li>Instant and professional cloning with consent and rights checks</li><li>A Pipecat service and a LiveKit Agents plugin</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Built around live phone calls and voice agents, so a project that only needs a narrated file may not need it</li></ul></div></div>

The latency figures in this pair can't be lined up, because each one is measured a different way, so the number a phone agent can plan around is the listening result, and on that Bland Speech v3 scored 1365 on 4 August 2026 while Rime wasn't tested. For phone calls and voice agents, that makes Bland Speech the one to try first. A team can read [Bland Speech vs Rime](/compare/bland-speech-vs-rime) and the [Rime review](/blog/rime-review) for Rime alone, the [Cartesia review](/blog/cartesia-review) for Cartesia, [text to speech compared](/compare/text-to-speech/rival-vs-rival) for more pairings, and the [text to speech hub](/compare/text-to-speech) for every comparison. Endpoints and sample code sit on the [text to speech API](/text-to-speech-api) page.

## Frequently asked questions

### Is Cartesia better than Rime?

Cartesia is better for many languages and quick voice cloning, and Rime is better for per-minute billing with a free monthly allowance and an enterprise VPC or on-prem option. Cartesia speaks 44 languages and clones from 10 seconds of audio, while Rime serves 10 languages on Coda and sells cloning as an enterprise order.

### Which is faster, Cartesia or Rime?

Rime publishes the lower numbers and Cartesia the simpler one. Cartesia says Sonic delivers first audio in under 90ms. Rime says Coda v2 has a 30ms median at the server without network time, and Rime's own benchmark gives Coda a 96ms median for one request and 150ms with 12 at once. Bland publishes 199ms for Bland Speech.

### Which is cheaper, Cartesia or Rime?

The two land close together once Rime's conversion of about 1,000 characters a minute is applied. Rime's Coda v2 is $0.050 a minute on Flex and $0.040 on Scale, close to $50 and $40 per million characters, and Cartesia's plans work out to $37.38 to $50 per million with the allowance used. Bland Speech is $40 per million characters.

### Which has more languages, Cartesia or Rime?

Cartesia has more. Sonic 3.6 speaks 44 languages, and Rime's Coda cloud API covers 10, with each voice serving one language.

### Can Cartesia and Rime clone a voice?

Both can. Cartesia clones an instant voice from 10 seconds of audio and trains professional voices on its Startup plan and above. Rime clones through an enterprise order, from 30 to 60 minutes of clean recordings for a basic clone.

### Are Cartesia and Rime HIPAA compliant?

Rime says it is HIPAA compliant, with its most recent audit completed in March 2026, and signs BAAs. Cartesia says it is SOC 2 Type II certified and offers BAAs on Enterprise. Bland is HIPAA compliant with a signed BAA.

### Which sounds more human, Cartesia or Rime?

Only Cartesia was scored. In Design Arena's Audio Realism Bench results of 4 August 2026, Cartesia's Sonic 3.5 ranked 6th at 1096, Rime's voices weren't in those results, and Bland Speech v3 ranked first at 1365.
