Kazakh speech to text API: Nova-3 and Sume language_code

Deepgram's changelog says Nova-3 added Kazakh on Sep 3, 2026. Sume STT 1.0 does not publish a language table, so for Kazakh audio send a public HTTPS audio_url and either omit language_code for auto-detect or pass a BCP-47 hint such as kk, then read the transcript yourself.

Deepgram facts are from its changelog; Sume's from the OpenAPI schema for POST /v1/stt-1.0/transcribe, read 2026-09-30.

What did Deepgram change?

The entry says Kazakh is a new Nova-3 language, alongside improved Nova-3 monolingual models for seven existing languages, for batch and streaming workloads. It describes Deepgram's service; it does not say anything about Sume.

What does Sume STT 1.0 accept?

Does Sume support Kazakh?

The docs do not list supported languages, and they do not name Kazakh. Do not assume it from Deepgram's changelog, because Sume's provider model ids stay internal. Run a short Kazakh sample and compare the text with what was said.

Should I set language_code or leave it out?

Try both on the same sample. Omitting it uses auto-detect; setting it gives the provider a hint. For the same trade-off in Korean, see STT language hint. Long files should go in clips of at most 10 minutes, submitted with mode: async as in Jobs and results.