Ten-prompt test grid on H3 Max: 5 s at 480p costs $3.125

Ten prompts, each a 5-second clip at 480p on MiniMax H3 Max, cost 10 x $0.3125 = $3.125 on Sume. Rerun the two winners at 1080p for 2 x $1.00 = $2.00 and the whole exercise lands at $5.125.

Five seconds is the shortest legal H3 Max clip, so it is the cheapest unit of test you can buy. Prices come from the Sume price tool (read 2026-10-09).

What the grid costs at each stage

The promote rows are alternatives, not a sum. For the all-in figure with a 1080p promote, add the grid and the 1080p row: $3.125 + $2.00 = $5.125.

The loop

Put the prompts in a file, one per line, and submit one async job each. The Idempotency-Key carries the batch name and the line number, so rerunning the script after a crash returns the original jobs and bills nothing new.

import os, requests
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
URL = "https://api.sume.com/v1/video-router/generate"
prompts = open("prompts.txt").read().splitlines()[:10]
for i, p in enumerate(prompts):
    body = {"model": "minimax-h3-max", "prompt": p, "resolution": "480p",
            "duration": 5, "aspect_ratio": "9:16", "mode": "async"}
    r = requests.post(URL, json=body, timeout=60,
                      headers={**H, "Idempotency-Key": f"grid-001-{i}"})
    r.raise_for_status()
    print(i, r.json())

Scale the grid

The grid is a per-clip price times the number of prompts. Five prompts is a quick check, twenty is a proper sweep.

Limits and gotchas

Each job reserves its full price at submit, so ten jobs reserve $3.125 together.

• Duration is 5-15 seconds on H3 Max; there is no 3- or 4-second grid.
• A prompt tested at 480p will not look identical at 1080p, which is a refinement from native 768p.
• Change a prompt and you must change its key. A key stands for one request body.
• The same rate applies at 480p on H3 and H3 Max, so the grid costs the same on either; pick the model by where the winner will end up.