AI video resolution price steps: where Sume cost jumps 2x or more

Resolution price is not a straight line. On Sume the biggest single step between neighbouring tiers is 3.33x, from 360p to 720p on gemini-omni-flash-1.1, and 3.0x from 768p to 2K on heygen-video-1. The pricier Wand Vega Pro rows jump about 2.5x from 1080p to 2K, then add only 10 to 14 percent for 4K.

The ratios below divide one tier's per-second rate by the tier beneath it, using the Sume video catalog (read 2026-10-12). Check GET /v1/videos/models, the live catalog, because rates change.

Which resolution step costs the most?

Divide each tier by the one before it. Examples: wan-3.0 0.125 / 0.0625 = 2.0 and 0.25 / 0.125 = 2.0. gemini-omni-flash-1.1 0.125 / 0.0375 = 3.33. wand-vega-1.5-pro 1.2250 / 0.4974 = 2.46. minimax-h3 0.1625 / 0.075 = 2.17.

Seedance rows are left out because they bill per 1,000 video tokens rather than by a per-resolution second rate, and kling-3 is flat at one 1080p rate.

What does a 10-second clip cost at the steepest step?

Multiply the rate difference by 10 seconds. wand-vega-1.5-pro from 1080p to 2K is (1.2250 - 0.4974) x 10 = $7.28 more, taking the clip from $4.97 to $12.25. gemini-omni-flash-1.1 from 360p to 720p is (0.125 - 0.0375) x 10 = $0.875 more, from $0.375 to $1.25. heygen-video-1 from 768p to 2K is (0.1125 - 0.0375) x 10 = $0.75 more.

The ratio and the dollars point at different rows. Heygen and Omni have the biggest multiples, but the Wand Vega Pro rows have the biggest dollar step.

When is the next tier above a steep step almost free?

After the steep step, several rows flatten. On wand-vega-1.0-pro 2K is $0.8269 and 4K is $0.9423, so 4K costs 14 percent more: 10 x (0.9423 - 0.8269) = $1.15. On minimax-h3 2K is $0.1625 and 4K is $0.20, so 4K adds $0.375 over 10 seconds. If you already decided to pay for 2K on these rows, check whether the 4K output is worth the extra cents.

How do I read the tier rates from code?

Fetch the catalog and print each model's pricing_skus map, then compute your own ratios. Field names follow the models endpoint response in the Sume docs.

import json, os, urllib.request

req = urllib.request.Request(
    "https://api.sume.com/v1/videos/models",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
)
data = json.load(urllib.request.urlopen(req))
for m in data["data"]:
    print(m["id"], m["pricing_skus"])

When is this the wrong approach?

Ratios alone can mislead.

• A big multiple on a small base can still be a few cents. Compare dollars per clip, not only ratios.
• Not every row offers every tier; the step only exists if the model lists both resolutions.
• Seedance token pricing depends on resolution through the token count, so use a token calculator, not this table.