How many 30-second jobs can you submit at once? Read generation_limits

Your in-flight budget is max(0, concurrency_limit - active_generation_jobs - queued_generation_jobs), capped at queue_capacity_remaining, all read from generation_limits on a submit response. A 30-second video holds a processing slot for minutes, so a batch bigger than that budget just queues behind itself.

Sume accepts valid jobs as queued while queue capacity remains, so full concurrency alone is not an error. The 429 queue_full error appears only when the queue is also full.

Plan limits

Processing concurrency comes from the plan, and prepaid top-ups do not raise it. The default queue capacity is max(3, concurrency_limit x 5). The dashboard Concurrency tab and the generation_limits.concurrency_limit field are the live source of truth, so prefer them to this table.

The wave submitter

This sketch submits a first job, reads its generation_limits, and returns how many more to send. Each new job counts against the budget until you read a fresh snapshot.

import os, requests
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}

def budget(limits: dict) -> int:
    free = limits["concurrency_limit"] - limits["active_generation_jobs"] \
           - limits["queued_generation_jobs"]
    return max(0, min(free, limits["queue_capacity_remaining"]))

def submit(prompt: str, key: str) -> dict:
    r = requests.post("https://api.sume.com/v1/video-router/generate",
        headers={**H, "Idempotency-Key": key}, timeout=30,
        json={"model": "wan-3.0", "prompt": prompt, "duration": 30,
              "resolution": "480p", "mode": "async"})
    r.raise_for_status()
    return r.json()

first = submit("Fog over a harbor at dawn", "harbor-001")
limits = first.get("generation_limits") or first["data"].get("generation_limits")
print("more jobs now:", budget(limits) if limits else "unknown, poll first")

When the snapshot is missing

The response includes generation_limits only when Sume can compute it. If it is absent, wait and refresh before you add more work. Never show wave_size_hint as your concurrency, because it counts queue slots too.