Python: quote one 10-second clip across five Sume video models

Before you run a side-by-side test of new video models, write down what the test will cost. This script prices one 10-second clip on five Sume ids from the list rates in the Video Router docs and pricing code, applies the 1.25 multiple, and rounds up to the cent per job.

Script

Rates are provider list dollars per second, except Seedance 2.5, which is priced by token. Resolutions differ: H3 and H3 Max use native 768p, not 720p.

import math

MARGIN = 1.25
PER_SECOND = {
    "gemini-omni-flash-1.1 720p": 0.10,
    "wan-3.0 720p": 0.10,
    "minimax-h3 768p": 0.06,
    "minimax-h3-max 768p": 0.08,
}


def billable(provider_usd):
    return math.ceil(provider_usd * MARGIN * 100) / 100


def seedance_25(seconds):
    tokens = 1280 * 720 * seconds * 24 / 1024
    return tokens / 1000 * 0.0214


seconds = 10
for name, rate in PER_SECOND.items():
    print(name, billable(rate * seconds))
print("seedance-2.5 720p", billable(seedance_25(seconds)))

What it prints

Using it

Floating-point rounding can nudge a figure by a cent in either direction; the poll response usage.cost is the authoritative number.

• Multiply by takes: three takes of each of five models at 10 s is about $30 here, and most of it is Seedance.
• Check GET /v1/video-router/models for rate changes; the docs date the Wan and H3 rows to fal's price pages.
• Run the real tests at 480p or 360p where a model allows it, and spend on a final only for the winner.