Five gateways resell the same Seedance 2.0 weights from ByteDance. Same model, same output. A five-second 720p clip on the Mini tier costs 0.18 USD on the cheapest of them and 0.77 on the dearest, which is 4.3 times the price for a file you could not tell apart.

That spread is not compute. Everyone is renting the same upstream capacity. It is markup, and it varies by tier in ways that are worth knowing before you wire one of them into a pipeline.

What five seconds actually costs

Matched test: 720p, 16:9, five seconds of output, no source video, in USD.

reAPI.ai is lowest on all three, by 15% against Replicate on Standard, 25% on Fast, and 58% against Atlas Cloud on Mini. The ordering behind us is not stable, which is the more useful observation: Atlas Cloud beats Replicate on Standard and loses to it on Mini, and WaveSpeed sits mid-pack on the cheap tier and near the top on the expensive one. Whoever was cheapest for your last project is not automatically cheapest for this one.

The lower rate that produces the bigger bill

Here is the part that catches people, and it has nothing to do with which platform you pick.

Every one of these gateways publishes a discounted per-second rate for requests that carry a source video. It looks like a reward for supplying reference footage. It is not. The discounted rate is charged against input seconds plus output seconds, not output alone, so the meter runs longer than the number you were quoting.

On reAPI.ai at 720p Standard, the two paths work out like this:

The second request has a rate 39% lower and a bill 22% higher. Any comparison table that multiplies the source-video rate by output duration alone will understate what you actually pay, and most of them do exactly that.

Two details worth pinning down, because they are the ones that surprise people:

Only a source video selects the lower rate. A reference image, a product shot, a first or last frame, an audio track: all of those stay on the base rate. And the source duration is rounded up to a whole second before it is added, so a 4.2-second clip bills as five.

With a source video attached, the ranking holds. A five-second source driving a five-second output at 720p is 0.938 on reAPI.ai against 1.10 on Replicate for the Standard tier, 0.680 against 0.85 on Fast, and 0.219 against roughly 0.512 on Atlas Cloud for Mini.

Picking a tier before picking a platform

The tier decision moves more money than the platform decision does, and it usually gets made last.

Mini to Standard is a 4.3x jump on the same platform. Across the five gateways in that table, the entire spread on the Standard tier is 1.97x. So a team defaulting to Standard on the cheapest gateway still pays more than a team running Mini on the most expensive one. If the output is short-form vertical video that gets watched once on a phone, the tier above Mini is frequently dead money.

Run one clip through Mini and Standard on the same prompt before committing. It takes ten minutes and it is worth more than any price table, including this one.

What reAPI is

reAPI.ai is a pay-as-you-go multimodal API. One key covers Seedance alongside other image, video and chat models, one credit is 0.001 USD, credits never expire, and a failed generation refunds in full automatically. No subscription, no minimum, and the per-tier rates plus the billing rules are printed on each model page rather than left to a support ticket.

Seedance 2.0 also runs at 4K there, which not every gateway on that list exposes.

The short version

At matched specs on Seedance 2.0, reAPI.ai was the lowest of these five on every tier, by 15% to 58%. But check the billing clock before you check the rate: on any of them, attaching a source video changes what gets counted, and that detail moves the bill further than the gap between two platforms.

Full tier tables are at reapi.ai/models/seedance-2-0 and reapi.ai/models/seedance-2-0-mini, with the source-video formula on the same page.