Cheapest AI video clip on each Sume model: the shortest legal length

The cheapest AI video clip you can submit through Sume's Video Router is a 3-second Gemini Omni Flash 1.1 clip at 360p, which bills $0.12, closely followed by a 2-second Wan 3.0 clip at 480p at $0.13 (read 2026-10-06). Every Seedance model starts at 4 seconds, MiniMax H3 and H3 Max start at 5 seconds, and Kling 3 starts at 4 seconds, so their cheapest clips cost $0.32 to $1.08.

A cheapest clip is useful for exactly one thing: proving that your integration works before you spend real money. It submits, reserves, renders and captures through the same path as a $5 clip, so one 12-cent job exercises your key, your webhook or poll loop, your storage code and your balance read. The table lists the shortest duration the Video Router accepts for each model, at the lowest resolution it offers (Kling 3 offers 720p and 1080p only).

Why the effective rate is above the list rate

Sume rounds each job up to a whole cent. On a 2-second Wan 3.0 clip at 480p the list rate is $0.05 per second, so the billable amount is $0.125, and the job bills $0.13. That is an effective $0.065 per second against a list-times-1.25 rate of $0.0625. The rounding is worth about 4 percent on that clip, and it shrinks as clips get longer, so 100 cheapest clips cost a little more than the rate card suggests.

Seedance is priced by video tokens, not by the second, so its effective rate depends on the resolution and the aspect ratio's pixel count rather than on a flat list rate. The Seedance rows above use 9:16.

• Wan 3.0 accepts 2 to 30 seconds, so it has the lowest floor.
• Omni accepts 3 to 10 seconds and has a 360p tier, which is why it wins on the very cheapest clip.
• H3 and H3 Max start at 5 seconds, and both bill the same at 480p.
• Read capabilities from the catalog for the live limits of each model.

What 100 cheapest clips cost

If you run the same smallest clip a hundred times, for a load test of your own queue code, the totals are $12.00 on Omni 360p, $13.00 on Wan 480p and $32.00 on H3 480p. The rounding shows up here: 100 Wan clips at the raw rate would be $12.50, but each bills $0.13, so the total is $13.00. A job that is too small to matter to you is still rounded to a cent, so batch your load test by job count rather than by seconds.

A smoke test that costs under a dollar

The script below prices a smoke test that submits the cheapest clip on four models, so you can see the total before you submit. The prices are the billed values from the table.

SMOKE = {
    "wan-3.0 480p 2s": 0.13,
    "omni 360p 3s": 0.12,
    "seedance-2-mini 480p 4s": 0.36,
    "minimax-h3 480p 5s": 0.32,
}

total_cents = sum(round(v * 100) for v in SMOKE.values())
print(f"smoke test: ${total_cents / 100:.2f}")
for name, price in sorted(SMOKE.items(), key=lambda kv: kv[1]):
    print(f"  {name}: ${price:.2f}")

What a cheapest clip will not tell you

A 360p clip is a poor judge of a 1080p brief. Resolution changes price and often changes what the model does with fine detail, so a cheap test says that the pipeline works and says little about the look. Use the cheapest clip for plumbing, then render the real brief at the resolution you will ship, one clip per candidate model, before you scale to a batch.

The model limits and prices are in the Video Router catalog and the video generation docs.