5-second AI video API: every Sume model that takes it, priced

Ten Sume video ids take a 5-second clip: Wan 3.0, Gemini Omni Flash 1.1, Kling 3, Grok Imagine Video 1.5, MiniMax H3, MiniMax H3 Max and all four Seedance ids. Five seconds is the shortest clip MiniMax sells, so it is the one length every one of them shares.

A 5-second job costs between $0.07 and $7.11 depending on the model and resolution. The cheap end is Grok at $0.07; the 1080p Seedance 2.5 clip is the expensive end.

What does a 5-second clip cost per model?

Billed price is the provider list times 1.25, rounded up to the cent per job. Seedance prices come from the repo's token estimate for a 16:9 clip without a reference video.

Why does Kling cost $1.05 in some requests?

The Kling 3 row lists one rate with audio off ($0.14 per second billed) and a higher one with audio on ($0.21 per second billed). A 5-second clip is $0.70 silent and $1.05 with generated audio. The catalog has no separate 1080p step for Kling.

Which 5-second row suits a hook test?

For a batch of hook variants, price matters more than peak quality, so start with the rows near the top of the table and move winners up. Ten 5-second hooks cost $3.20 at 480p on Wan 3.0, $6.30 at 720p, and $0.70 on Grok if you already have the starting stills.

Rows that take references (Wan, Seedance, MiniMax) keep a product or character consistent across the batch. Kling 3 and Grok do not take reference inputs.

How do you submit it?

Pin the model and send duration: 5. Read the key from the environment, and add an Idempotency-Key so a retry returns the original job instead of a second charge.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/videos",
    headers={
        "Authorization": f"Bearer {os.environ['SUME_API_KEY']}",
        "Idempotency-Key": "hook-001",
    },
    json={
        "model": "wan-3.0",
        "prompt": "Hand drops a red shoe onto a white table, 9:16",
        "duration": 5,
        "resolution": "480p",
        "aspect_ratio": "9:16",
    },
)
print(r.status_code, r.json()["id"])

What is the cheapest way to confirm the numbers?

Submit one 5-second job at the lowest tier and read usage.cost in the poll response. It is the Sume billable amount, so it matches the table. GET /v1/videos/models returns each model's pricing_skus if you want the rate before you spend anything.