Estimate a Sume video request first: rate x seconds x 1.25

You can estimate most per-second video requests before you send them: take the provider list price per second, multiply by the seconds you request, and add the 25 percent margin. The video generation docs state that Sume reserves the amount at submit and that the actual amount appears in usage.cost. This post gives the formula and a tiny script.

Models priced per token, such as the Seedance family, do not fit this formula; read their rates from the catalog instead.

Worked examples

A small estimator

The code below applies the margin in integer micro-dollars, rounding up, as the pricing package does: the margin is 125/100 and the result is rounded up to the next micro-dollar. Run it with Python 3.

from decimal import Decimal, ROUND_CEILING

RATES = {  # provider list price per second, read 2026-10-05
    "wan-3.0-720p": Decimal("0.10"),
    "kling-3-pro-audio-on": Decimal("0.168"),
    "omni-flash-1.1-1080p": Decimal("0.15"),
}

def sume_price(list_usd: Decimal) -> Decimal:
    micros = list_usd * 1_000_000
    billed = (micros * 125 / 100).to_integral_value(rounding=ROUND_CEILING)
    return billed / 1_000_000

for name, rate in RATES.items():
    print(name, sume_price(rate * 10))

Reserve, then settle

Think of the estimate as the reserve. If the request completes as asked, the settled usage.cost should match your arithmetic; if it differs, trust usage.cost, since that is the billable amount. Log both numbers per request and the gap tells you whether your assumptions about duration and resolution hold.

Keep enough balance for the reserve of every request in flight. Reading the balance and usage endpoints is covered in the billing and credits docs. The rate card on the Sume API rate card lists fixed-price tools such as TTS and captions that need no formula.

• Estimate = list per second x seconds x 1.25.
• Compare estimate with usage.cost for every call.
• Use the catalog for token-priced models.