Inline vs Queued + Polling

How a request behaves depends on length: short text comes back immediately; long text becomes a job you poll.

At a glance

Short text → inline audio in the response.
Long text → a job id + poll URL.
Poll until complete, then fetch the audio.

Inline

Below the inline threshold, synthesis returns audio directly. One request, one response, simplest for short snippets.

Queued + polling

Above the threshold, you get a job id and a poll URL. Poll GET /api/v1/tts/jobs/{id} until it's done, then download from /jobs/{id}/audio.

POST /api/v1/tts/synthesize  -> { "job_id": "...", "poll": "/api/v1/tts/jobs/..." }
GET  /api/v1/tts/jobs/{id}    -> { "status": "running" | "done" | "failed" }
GET  /api/v1/tts/jobs/{id}/audio -> the audio file