Lyria 3.5 has no Batch API: queue many tracks on Sume instead

Google's Lyria 3.5 model page says batch, Flex and Priority inference are not supported (read 2026-10-01). On Sume you can still submit many music jobs: valid jobs are accepted as queued when your workspace is at its processing limit, and run as slots free up.

How does queueing work?

Sume's generation admission docs say concurrency is a dispatch limit, not a submit limit. Processing concurrency is plan-based: Free 1, Pro 4, Startup 8, Scale 20. Queue capacity defaults to max(3, concurrency x 5). A full queue returns 429 queue_full, and too many requests returns 429 rate_limited.

What does each side document?

How do I submit a list?

One POST per track, each with its own Idempotency-Key, then store the job ids. Do not treat queued as a failure.

import os
import requests

H = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
names = ["Maya", "Leo", "Sam"]
jobs = []
for i, n in enumerate(names):
    r = requests.post(
        "https://api.sume.com/v1/music-router/generate",
        headers={**H, "Idempotency-Key": f"bday-batch-{i}"},
        json={"model": "lyria-3.5", "prompt": f"Birthday pop song for {n}"},
    )
    r.raise_for_status()
    jobs.append(r.json()["id"])
print(jobs)

Does Sume price differ for batch-style use?

Sume's price is a fixed $0.125 per generation with no batch discount in the docs. Google's page does not give a Sume-comparable number here, so do not assume a saving either way.