# Seedance API pricing: cost per second for every model and resolution By SeedRouter · Published 2026-09-25 · Updated 2026-09-25 Seedance API pricing on SeedRouter is per second of video, and the price depends on two things: the model and the output resolution. Seedance 2.0 Mini is the cheapest per second, Seedance 2.0 Fast sits in the middle, Seedance 2.0 costs more and adds 1080p and 4K, and Seedance 2.5 has the highest rate and the longest clips. There is no subscription, and failed tasks are not charged. Under the hood every Seedance model bills video tokens, the official unit, and each resolution produces a fixed number of tokens per second. The tables below turn the live token rates into a price per second, for all four models. ## What does one second of Seedance video cost right now? ### Seedance 2.0 Current rates are temporarily unavailable. No price estimate is provided; unavailable rates must not be interpreted as free usage. ### Seedance 2.0 Fast Current rates are temporarily unavailable. No price estimate is provided; unavailable rates must not be interpreted as free usage. ### Seedance 2.0 Mini Current rates are temporarily unavailable. No price estimate is provided; unavailable rates must not be interpreted as free usage. ### Seedance 2.5 Current rates are temporarily unavailable. No price estimate is provided; unavailable rates must not be interpreted as free usage. Each model has one table. "USD per second" is the price of one second of 16:9 video without a reference video; the column next to it is the same second when the request includes a reference video. The last two columns are the raw rates per million video tokens. The figures are read from the same rate table that bills your account; if a rate cannot be loaded, the table says so instead of showing zero, and you should treat it as unknown, never as free. ## How are Seedance video tokens counted? The token count follows the size and length of the video: ```text video tokens = (output seconds + reference video seconds) × width × height × 24 / 1024 cost = video tokens × rate per 1M tokens / 1,000,000 ``` At 16:9 that gives a fixed count per second of output: | Resolution | Seedance 2.0, 2.0 Fast, 2.0 Mini | Seedance 2.5 | | ---------- | -------------------------------- | ------------ | | 480p | 10,044 | 9,607.5 | | 720p | 21,600 | 21,600 | | 1080p | 48,600 | 48,600 | | 4K | 194,400 (Seedance 2.0 only) | not offered | Text prompts, reference images and reference audio are not billed. Other aspect ratios have slightly different pixel counts, so their price per second differs a little from the 16:9 figures. The charge follows the tokens the finished video reports in `usage.completion_tokens`, not the number you asked for. Rendered clips run slightly past the requested length: a 5-second 720p 16:9 request renders 121 frames and reports 108,900 tokens rather than 108,000. With `duration: -1` the model picks the length, and you pay for the length it actually made. ## Does a reference video make a request cheaper? The rate is lower, but the tokens are not. A request that includes a reference video uses the reference rate for all of its tokens, and the reference video's own seconds are added to the count. With a reference clip as long as the output, the token count doubles. So the rule is simple: an equally long reference video saves money only if the reference rate is less than half of the plain rate. Compare the two "USD per second" columns in the table above for your model and resolution. A short reference clip under a long output is where the lower rate helps most. One more detail from ByteDance's [ModelArk pricing page](https://docs.byteplus.com/en/docs/ModelArk/1544106): "When the input includes video, minimum token consumption limits apply." The charge is always the token count the finished video reports, so a very short reference clip can be billed at that minimum rather than at its own length. ## How do you estimate a whole clip? Multiply the price per second from the table by the length: 1. Pick the model and resolution row. 2. Take "USD per second", or the reference column if the request has a reference video. 3. Multiply by the output seconds. With a reference video, multiply by output seconds plus reference seconds. A minute of footage is 60 of those seconds. Seedance 2.0 and its Fast and Mini versions make up to 15 seconds per task and Seedance 2.5 up to 30, so a minute is four or two tasks. The calculators on the [Seedance 2.0](https://seedrouter.ai/models/seedance-2-0#pricing), [Seedance 2.0 Fast](https://seedrouter.ai/models/seedance-2-0-fast#pricing), [Seedance 2.0 Mini](https://seedrouter.ai/models/seedance-2-0-mini#pricing) and [Seedance 2.5](https://seedrouter.ai/models/seedance-2-5#pricing) pages do the same arithmetic for the settings you pick. ## What is the cheapest way to run Seedance? | Goal | Cheapest option | Trade-off | | ---------------------------- | -------------------------------------- | ------------------------------------------------ | | Lowest cost per second | Seedance 2.0 Mini at 480p | Stops at 720p | | Drafts before a final render | Mini or Fast at 480p, short `duration` | Re-render the keeper at the final settings | | 1080p output | Seedance 2.0 | Seedance 2.5 also renders 1080p at a higher rate | | 4K output | Seedance 2.0 | The only model with 4K | | Clips longer than 15 seconds | Seedance 2.5 | Highest rate per second | All three Seedance 2.0 models accept the same request body, so moving a job from Seedance 2.0 to Fast or Mini is a one-word change to `model`. The [Seedance 2.0 vs Fast vs Mini guide](https://seedrouter.ai/blog/seedance-2-0-vs-fast-vs-mini) covers what else changes. ## When is a Seedance task not charged? A task that ends `failed` or `expired` is not charged. That includes media that breaks a model limit, which is checked when the task starts and fails before any generation, and output withheld by content filtering. A polling timeout on your side is not a failure: keep the task ID and check the same task again, because a second submission is a second video to pay for. ## Frequently asked questions ### How much does Seedance 2.5 cost? Seedance 2.5 bills the same video tokens as Seedance 2.0 at a higher rate, and the table above shows its live price per second at 480p, 720p and 1080p. One task can run up to 30 seconds, so budget per task by multiplying that price by the length you request. ### Is there a Seedance subscription? Not on SeedRouter. You top up once on the [billing page](https://seedrouter.ai/billing), the credits never expire, and each finished video is deducted from that balance. The same balance pays for every model in the catalogue. ### Why is Seedance 2.5 more expensive than Seedance 2.0? Its rate per video token is higher. At the same resolution and length a Seedance 2.5 clip has about the same number of tokens as a Seedance 2.0 clip, so the whole difference is in the rate. What it buys is longer clips, up to 30 image references, and editing or extending a video you supply; the [Seedance 2.5 vs 2.0 guide](https://seedrouter.ai/blog/seedance-2-5-vs-2-0) compares them. ### Where can I see what a video cost? In [usage history](https://seedrouter.ai/usage). The task response reports `usage.completion_tokens`, the billed quantity, but it carries no dollar amount. ## Price it before you send it Pick the model by the output you need, draft cheaply at 480p on Mini or Fast, and render the final once. Check the live tables above before a large batch, top up on the [billing page](https://seedrouter.ai/billing), and see the [Seedance 2.0](https://seedrouter.ai/docs/seedance-2-0) and [Seedance 2.5](https://seedrouter.ai/docs/seedance-2-5) API references for every request field.