8-second 720p Seedance price ladder: mini, fast, 2.0 and 2.5 on Sume

An 8-second 720p 16:9 clip prices at about $1.51 on seedance-2-mini, $2.42 on seedance-2-fast, $3.02 on seedance-2 and $4.62 on seedance-2.5. Those figures come from Sume's token formula for Seedance and the documented rule that Sume bills the provider list times 1.25 on video models. They are estimates; the job's usage.cost is the number that counts.

OpenAI's Sora Videos API ended on 2026-09-24 with no replacement named, according to the Pondero report. Seedance is one of the alternatives that report lists as having current API availability, and it comes in four tiers on Sume, so the first decision is which tier.

The formula

Seedance prices in video tokens, not seconds. In Sume's pricing code a clip is width x height x seconds x 24 / 1024 tokens, rounded up, at 24 frames per second. 720p at 16:9 is 1280 x 720, so 8 seconds is 172,800 tokens. 9:16 is 720 x 1280, the same pixel count, so the same tokens.

Each tier has its own list rate per 1,000 tokens. Multiply, then multiply by 1.25 for the Sume price. The documented duration range is 4 to 15 seconds for the 2.0 family and 4 to 30 seconds for seedance-2.5.

Reading the ladder

The gap between the cheapest and dearest tier is $4.62 against $1.51, about 3.1x for identical pixels and length. seedance-2-mini is the draft tier: if you are rewriting prompts that were written for a different model, run the whole set there first and only re-render the keepers up the ladder.

Tier order is not a quality promise in the docs. The pricing page tells you what each costs; the only way to learn which one matches your old Sora look is to render the same prompt on two of them and compare.

Things the table leaves out

Resolution moves the price a lot. 1080p has 2.25 times the tokens of 720p. seedance-2.5 also carries a higher per-token rate at 1080p, covered in a separate post. Reference videos add a surcharge, covered here.

Check the live catalog before you hard-code any of this. GET /v1/videos/models lists pricing_skus per model, and the docs state that limits differ for each model.

A one-line sanity check

Before a large batch, compute the expected reserve yourself: tokens x list rate / 1000 x 1.25. If the number on a test job's usage.cost is far from it, one of your parameters (resolution, aspect ratio, a reference video) is not what you thought.

Using the ladder to pick a tier

A common pattern for a migration is two passes. Pass one renders every prompt on seedance-2-mini at 720p to check framing, motion and whether the prompt survives the move from a different model's style. Pass two re-renders the ones you keep on seedance-2 or seedance-2.5. For 100 prompts that is roughly $151 for the draft pass at the table price, then the per-keeper cost of the higher tier for however many survive.

The same ladder works as a fallback order. If your first-choice model returns a failure, you can step down to a cheaper tier with the same duration range, since all four Seedance ids accept 4 to 15 seconds. Keep one Idempotency-Key per model attempt so that a retry of the same attempt is safe.