2026-10-29 · VideoFreeTier

Seedance 2.5 API Free: the 2M Token Headline Sits Above a Row That Says None

There is no free Seedance 2.5 allowance on BytePlus ModelArk. The product page advertises "Free quota — Up to 2M free tokens per model" near the top. Scroll to the pricing block on the same page, open its Free tokens tab, and the Dreamina-Seedance-2.5 row in the Vision models table holds a single slash. The two 2,000k token grants in that table belong to Seedance 1.5 Pro and Seedance 1.0 Pro — the generation before this one.

Everything below was read on 29 October 2026 from byteplus.com and ai.byteplus.com as a guest, with no account, no API key and nothing spent. The Free tokens tab was clicked rather than inferred from the collapsed pricing block, and every figure was cross-read a second time from the Markdown versions of the documentation pages, which the site serves next to the rendered ones.

What the page says first

The ModelArk product page puts four link cards under its hero. The first is labelled Free quota, carries a Hot badge, and its subtitle reads "Up to 2M free tokens per model". Nothing on that card narrows the claim to a model family, a region or an account type.

Further down the same page the claim gets a qualifier: "Get started with 500,000 free tokens for every Large Language Model and 2,000,000 free tokens for every vision model."

That sentence is where the 2M comes from, and it is a per-model grant scoped to the words "vision model". Seedance 2.5 sits inside that bucket on this same page: the model card at the top of the listing tags it Vision models and then Video generation. The page's own taxonomy files a video model under the category the 2M sentence covers.

Which is why the next part of the page matters more than the headline.

The Free tokens tab, and what the 2.5 row holds

The pricing block under the model list has two tabs, Standard and Free tokens. The second is the only place on the public site that answers the question model by model, and it opens without a login.

Under Vision models the table has two columns, Model name and Quantity. Read down it:

The slash is the page's way of writing "nothing here". It is not a zero and it is not a dash: every row that has an amount prints one, and every row that has none prints the slash.

So the 2,000k grants are real, and they are attached to Seedance 1.5 Pro and Seedance 1.0 Pro. The model most people reading the 2M headline are asking about is not one of the two that receives it.

The same answer, from three other places on the site

The Standard tab next to it makes the point from the other direction. It prices Seedance 2.5 at $6.4 per million tokens with video input and $10.7 without. The image models get a free unit written into the rate itself — Seedream 5.0 Pro's input line reads "$0.0030USD/piece input diagram (The first one is free)". The video rows carry rates and nothing else.

The documentation repeats the shape. The pricing page's video generation section has no free-quota column anywhere; the only Free quota column in that entire document belongs to the embedding section, where skylark-embedding-vision is listed at 500K tokens. The model list's video generation limits table has three columns — max RPM, max concurrency, Flex support — and no free-quota column, while the 3D generation limits table two screens further down carries one.

Four places, one answer. The number exists on this platform, and it is not attached to this model.

Activating the model is where "free" stops

Free quota aside, Seedance 2.5 cannot be called at all until the model is activated, and the activation threshold is the part the headline does not mention.

The activation document lists three routes, any one of which is enough. Credits: a verified Credits user whose Credits reach USD 30. AI Savings Plan: a dedicated plan of USD 30 or more with a remaining balance. Resource pack: packs worth USD 30 or more with remaining quota.

The tutorial page says the same thing in its opening note, under the words "make sure you meet one of the following conditions": account balance above USD 30, a dedicated AI Savings Plan at the USD 30 tier or above, or a Seedance 2.5 resource pack with available quota.

Four steps later, on the same page, the quick-start checklist narrows it: "Activate the model: Make sure you have purchased a resource pack. Otherwise, you cannot activate Seedance 2.5."

And the resource pack rules page — the document the other two link to for the details — lists activation as four options, all of them packs: a Seedance 2.0 pack, a 2.0 fast pack, a 2.0 mini pack, or a 2.5 pack, each with remaining quota. Credits and the Savings Plan do not appear on that list at all.

Three pages, three scopes: three routes, three routes, then packs only. The narrowest version is the checklist a first-time API user is told to follow, and it is written as an absolute.

For anyone costing this out: the smallest Seedance 2.5 pack is 5 million tokens at USD 32, valid 90 days, non-refundable, and it offsets only Seedance 2.5 tokens. The 10-million pack is USD 64 and the 100-million pack is USD 640. A 5-million pack works out at roughly 62 five-second 480p clips once the page's own deduction ratio is applied — our arithmetic, not a published figure.

The billing source table, which is one line long

The resource pack rules page closes its activation section with a table headed Available billing sources. It is the cleanest single answer the site gives.

For Seedance 2.5 the entry reads: "Seedance 2.5 resource pack remaining quota / BytePlus account balance."

Two sources. The new-user free quota is not one of them. The AI Savings Plan is not one of them either, which is a second place where these pages disagree with each other, since the activation document puts the plan second in its deduction order.

The 500k mode that cannot be used here

There is a second free mechanism on the platform, and it is the one that sounds like it should apply. Free Tokens Only mode lets an account call inference and consume only the 500k tokens the platform grants, pausing the service before anything is billed.

On the model side its scope is broad: "All models are supported except for Multimodal embedding (vectorization) models and Simultaneous interpretation models." A video generation model is not excluded by that sentence.

On the account side it closes. Eligibility is "accounts that have not activated model services and have never enabled or disabled Free Tokens Only mode."

Seedance 2.5 requires activation. Activation is exactly what removes the eligibility. The two conditions cannot both be true, so the one free path that would let a call actually run is closed to the model that needs activating — unless the mode is switched on in the same step as activation, which the documentation offers as an option on the activation page without reconciling it with the eligibility sentence.

Whether that combination works in practice was not tested here, because it needs an account. It is the largest gap in this page.

What 2M tokens would be worth, if it applied

The arithmetic is worth running even though the row holds a slash, because the headline is the number people plan around. The pricing page publishes the formula: token consumption is input duration plus output duration, multiplied by width, by height, by frame rate, divided by 1024.

Run that for a five-second 16:9 clip at 24 fps and 2,000,000 tokens comes to about 41.6 clips at 480p, 18.5 at 720p and 8.2 at 1080p. At the 500k figure it would be 10.4, 4.6 and 2.1.

That is our arithmetic, not a published figure, and it reconciles with the vendor's own price examples: the same formula reproduces $0.514, $1.156 and $2.843 for a five-second clip at the three resolutions, to the cent. The formula is sound. The grant is what is missing.

What people are actually asking

The question turns up in comments under Seedance 2.5 comparison videos, and it is usually not framed in tokens:

"Is this paid or is this free. If its paid then why to get in to that all APi thing, just top up and generate... Plz explain" — @@amanthakur2264, YouTube comment, collected 29 October 2026

Another comment in the same thread reports a call refused for insufficient credits. Both are user reports, not measurements, and they are recorded as reports. They describe the same wall the documents describe: the documented steps for creating an API key are four console actions and none of them is a payment, but none of them activates a model either.

What this does not settle

No account was created, no key was generated and no call was made. Whether a new account's console shows a free quota against Seedance 2.5 that the public table does not is not known here — the documentation points at the Model Activation page for per-account quota, and that page needs a login.

The 2,000k rows for Seedance 1.5 Pro and 1.0 Pro were read, not spent, so what those tokens buy in practice is untested too.

None of these pages carries a visible revision date; the tutorial's footer and the product page's pricing block both come up without one. What is written here is the state of those pages on 29 October 2026, and the single cell worth re-checking later is the Quantity column on the Dreamina-Seedance-2.5 row.

Related on this site: How much Seedance 2.5 you actually get for free — the consumer channels, where the allowance is a daily credit top-up rather than an API grant. Seedance 2.5 credits to dollars — a channel that publishes both a credit cost per second and a price for credits, which is how a five-second clip gets priced. CapCut Seedance 2.5 free — the same question asked of an editing app rather than an API, with the same shape of answer.