Rime Arcana retired: use Coda; Sume TTS Router model_not_found

Rime retired Arcana from its cloud API on 2026-08-15: requests using arcana, arcanav2 or arcanav3 are now served by Coda, and Rime says to replace the retired identifier and verify the voice and language pairing. Sume's TTS Router takes the opposite approach to unknown ids: it fails with 400 model_not_found and returns the catalog URL.

Rime facts are from its changelog; Sume facts from the OpenAPI reference, read 2026-10-01.

What should I change on the Rime side?

Per the changelog, old Arcana requests keep working because Coda serves them, but the identifier is retired. Update your code to the Coda model id and check that each voice and language pairing still holds, since the model behind the id changed.

What does Sume's TTS Router do with an unknown id?

The router requires an explicit model, an id from GET /v1/tts-router/models. An id that is not in the catalog is rejected with 400 model_not_found; nothing is remapped. The catalog description says v1 is Sonic only, and the enum includes sonic-3.6. Authenticate with your Sume API key only.

How do the two behaviors compare?

How do I avoid a hard failure in production?

Fetch the catalog at deploy time or on a schedule, and keep the model id in config rather than code. A loud 400 is easy to alert on; a silent remap changes audio without a code change. Note that plain TTS 1.0 has no engine picker, so the model id only applies to the router. See also the Turbo v2.5 deprecation post.