n above the ceiling returns 400 on Sume images: split the batch

If you send n: 5 to Seedream 4.5 on Sume's image API you get a 400 that names the allowed range, because the request schema accepts n up to 10 but each model publishes a lower ceiling: 4 on most models, 1 on Grok Image. The ceiling is the n range descriptor in GET /v1/images/models. The Image API docs call this out: use n for up to 10 images, but per-model ceilings are lower.

The fix is not a retry. Read the ceiling once, then split the batch.

Where do I read the ceiling?

Each catalog row lists supported_parameters, and n is a range descriptor with min and max. For most models that reads { "type": "range", "min": 1, "max": 4 }. Sume does not clamp or drop a value outside the range; it rejects the request, which is why the error is safe and cheap to hit.

How do I split a batch?

Plan the call sizes first, then submit them. The helper below turns a total and a ceiling into call sizes and runs offline, so you can test it before any key is involved.

def plan(total: int, cap: int = 4) -> list[int]:
    sizes = []
    while total > 0:
        sizes.append(min(cap, total))
        total -= cap
    return sizes

assert plan(10) == [4, 4, 2]
assert plan(6, 1) == [1, 1, 1, 1, 1, 1]
assert plan(3) == [3]
print(plan(10))

# For each size, POST /v1/images with {"n": size, ...}.
# Price is cost_usd x n either way, so splitting costs nothing extra.

Does splitting change the bill?

No. Sume charges cost_usd x n, so ten images are ten times the per-image price whether you send three calls or ten. What changes is failure shape: each call is all-or-nothing, so a smaller call that fails wastes less time. Give each call its own Idempotency-Key, derived from your item id and the call index, so a retry returns the original job. Prices here are list x 1.25 in dollars per image before whole-cent rounding; the catalog's billable formula reads "list x 1.25, ceil usd cents", so confirm the first charge in usage.cost and budget from the invoice, not from the sum.