Kling 4.0 Flash 3-20 s at 720p: the closest Sume models, priced

Kling 4.0 Flash is 3 to 20 seconds at 720p, so the closest Sume models are Wan 3.0 (2 to 30 seconds), Gemini Omni Flash 1.1 (3 to 10), Seedance 2.5 (4 to 30) and Kling 3 (4 to 15). At 720p, a 10 second clip is $1.25 on Wan 3.0 and on Omni, $1.40 on Kling 3 with audio off, and $5.78 on Seedance 2.5. Only Wan 3.0 and Seedance 2.5 reach 20 seconds.

The details of Flash come from the fal explainer (read 2026-10-05): limited early access since the 2026-09-28 announcement, 720p only, 8-bit SDR output. No Flash price is given there, so this post compares what Sume lists, not the cost of Flash.

Floors, ceilings and rates

Each model has a different floor and ceiling. The Sume video docs list Wan 3.0 at 2 to 30 seconds, Seedance 2.5 at 4 to 30, Kling 3 at 4 to 15, and Gemini Omni Flash 1.1 at 3 to 10 seconds. A Flash-style request at 3 seconds therefore fits Wan 3.0 and Omni, but not Seedance 2.5 or Kling 3, which start at 4.

Wan 3.0 and Omni are billed per second at 720p, $0.125 each. Kling 3 is $0.14 per second with audio off and $0.21 with audio on. Seedance 2.5 is billed by video tokens, which at 720p 16:9 come to about $0.58 per second.

The gap between the cheapest and the dearest is large. At 10 seconds and 720p, Seedance 2.5 is about 4.6 times the price of Wan 3.0, so the choice of model matters more to the bill than the length does. If you only need a Flash-style test clip, begin with the lower-priced models and move up only if the result is not good enough.

720p clips at Flash lengths

Which one to pick

Read the table by use. For a 3 to 10 second clip with sound, Omni is the one that always has sound, so it matches Flash most closely on length and price. For 20 seconds at 720p, Wan 3.0 is the cheapest at $2.50, and Seedance 2.5 is the one that adds 21:9 and the widest reference support.

Sound differs between them. Omni's audio is always on, Kling 3 and Seedance 2.5 let you choose, and the Sume docs do not describe a separate audio switch for Wan 3.0 in the pages read for this post, so test it before relying on it.

Wan 3.0, Seedance 2.5 and Kling 3 list an end frame, which lets you pin how a clip finishes. Omni is the odd one out for ratios, since it takes only 16:9 or 9:16, so a 21:9 or 1:1 frame has to be cropped afterwards. Cropping with the video filter is a separate step and costs $0.02 per encode.

The price table in Python

This prints the table from the rates above, so you can change the lengths. It does not call the API.

WAN_OMNI = 0.125
KLING = {"off": 0.14, "on": 0.21}
SEEDANCE_720P_PER_SECOND = 1280 * 720 * 24 / 1024 / 1000 * 0.0214 * 1.25

for s in (3, 5, 10, 15, 20):
    wan = round(s * WAN_OMNI, 2)
    omni = round(s * WAN_OMNI, 2) if 3 <= s <= 10 else None
    kling = (round(s * KLING["off"], 2), round(s * KLING["on"], 2)) if 4 <= s <= 15 else None
    seed = round(s * SEEDANCE_720P_PER_SECOND, 2) if s >= 4 else None
    print(s, wan, omni, kling, seed)

Then submit

Submit the one you pick to POST /v1/videos with the length, resolution of 720p, and an Idempotency-Key. Read the allowed values from GET /v1/videos/models first, as the video docs advise, because the catalog is the source for ranges. If Kling 4.0 appears in that list, its entry will show its own durations, ratios and prices.

Remember that a price in this table is a reserve at submit, and the billed amount is usage.cost. For a three-second clip on Omni the reserve is $0.38, which is small enough to test each of the models on the same prompt, at the same length where their ranges overlap, and to compare them by eye before you commit to one.