Cartesia Sonic 3.6: 44 languages or 61 locales, which to quote

Quote 44 languages if you mean languages, and 61 locales if you mean language-and-region variants; Cartesia uses both numbers on its own pages. On Sume you reach Sonic through the TTS Router, where the request carries a BCP-47 language code and the model id is sonic-3.6, sonic-3.5, sonic-3, sonic-latest or sonic-preview.

The two numbers are not a contradiction, but they answer different questions, and mixing them up in a spec sheet or a client deck is an easy way to promise a language you cannot deliver. This post lays out where each number comes from and how to turn it into something testable on Sume.

Where each number comes from

The Sonic product page says 44 languages. The Sonic 3.6 launch post talks about 61 locales and 500+ preset voices, and singles out Odia and Urdu as new, both above 96% transcript accuracy, plus Hinglish code-switching. The latest-models doc lists Sonic 3.6 as generally available at 44 languages, adds Odia and Urdu, and says it is compatible with 3.5.

A locale is a language as spoken in a region, so one language can account for several locales. Read together, the pages are consistent: 44 is the language count and 61 is the locale count. When you write copy, pick the one that matches your claim and name the unit.

What the Sume request actually takes

The TTS Router contract takes a required model id and an optional language. The language field is a BCP-47 string of 2 to 16 characters, and for any language other than English you should set it explicitly so the job does not guess. The catalog on Sume is Sonic only, so Cartesia's language list is the list that matters here.

Sume does not publish its own language count for the router. It passes the code through to the model. That means the only honest count for your project is the set of languages you have tested with your own script and voice.

• Set language on every non-English request.
• Keep the voice and the language consistent; the voice-language mismatch warning exists to catch a voice speaking a language it was not built for.
• Use sonic-3.6 or the sonic-latest alias, which points at 3.6, once you have confirmed the audio.

Turn the claim into a test

Pick the five languages and locales your customers use, write one 20-second script for each, and run them through the router with a fixed voice. Listen for dropped words, wrong stress, and numbers read in the wrong language. See how to audition one voice across languages for a test matrix.

Record the model id from the job result next to each clip. TTS jobs on Sume record the model, voice, language and output format, so the test is reproducible later when a new snapshot ships.

Wording that stays accurate

For a deck: Sonic 3.6 supports 44 languages and 61 locales according to Cartesia (cite both pages). For a Sume spec: pass a BCP-47 language code; the model decides what it can speak. Avoid stating that Sume supports 61 of anything, because that is a vendor figure, not a Sume guarantee.