GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone
Seedance 2.5 Pricing: 53% More Per Token Than Seedance 2.0
2026/08/05

Seedance 2.5 Pricing: 53% More Per Token Than Seedance 2.0

Seedance 2.5 pricing is $10.70 per million tokens on BytePlus. See the per-second conversion, reference-video clock, and current reAPI rates.

Seedance 2.5 pricing is live in ByteDance's docs at $10.70 per million tokens without video input, against $7.00 for Seedance 2.0[1]. The model is also callable on reAPI as doubao-seedance-2.5-face.[6] The official token rate is 52.9% higher than 2.0 for the same two resolutions.

The per-token framing matters more than it sounds, because AI video is quoted per second and metered per token. The gap between those two units is where most cost surprises live, and Seedance 2.5 widens it in a way that breaks any conversion factor you built for 2.0.

TL;DR

  • Seedance 2.5 costs $10.70 per million tokens without video input and $6.40 with it, against $7.00 and $4.30 for Seedance 2.0 — 52.9% and 48.8% increases[1].
  • Only 480p and 720p are published for 2.5. The current production route does not accept 1080p or 4K[1][6].
  • Billing runs on (input_video_seconds + output_seconds) × width × height × fps / 1024, with fps fixed at 24[1]. Your reference clip is metered exactly like generated frames.
  • 480p changed frames between versions. Back-solving ByteDance's own worked examples puts 2.5 at ~854×480 and 2.0 at a taller frame, which is why 480p rises 46% per second while the token rate rises 53%.
  • 2.5 accepts up to 30 seconds of reference video, double 2.0's 15. A five-second 720p job runs $1.244 with a short reference and $4.838 with a 30-second one[1].
  • Seedance 2.0 on reAPI runs $0.024 to $0.780 per billed second depending on variant, resolution, and source-video mode[2].

Seedance 2.5 pricing, and the rate that looks wrong

ByteDance publishes token rates per model and input mode[1]:

ModelNo video inputWith video input
Seedance 2.5 (480p, 720p)10.706.40
Seedance 2.0 (480p, 720p)7.004.30
Seedance 2.0 (1080p)7.704.70
Seedance 2.0 (4K)4.002.40

Seedance 2.5 occupies a single row because it ships with a single rate. Read the 4K row underneath it again, though: it is the cheapest tier per token by a wide margin, 43% below 480p.

It is also, by a wide margin, the most expensive output you can buy. A 3840×2160 frame carries 19.4 times the pixels of what 480p actually renders. The rate falls 43% while the token count rises 1940%. Anyone comparing providers by scanning the rate column will conclude 4K is a bargain, and will be wrong by a factor of eleven.

This is the recurring hazard with token-metered video. The number on the price sheet is only half of the multiplication.

Where the tokens come from

The formula ByteDance documents is[1]:

tokens = (input_video_seconds + output_seconds) × width × height × fps / 1024

Frame rate is fixed at 24. Pixel dimensions per label:

LabelPixelsTokens per second
480p (Seedance 2.5)~854 × 4809,607
480p (Seedance 2.0)864 × 49610,044
720p1280 × 72021,600
1080p1920 × 108048,600
4K3840 × 2160194,400

Checking the formula against a published example: five seconds of 720p on 2.0 with no reference is 5 × 21,600 = 108,000 tokens, times $7.00 per million, or $0.756. ByteDance's own worked example for that exact configuration reads $0.756[1]. The formula is exact, not an approximation.

The 480p frame changed and nobody announced it

This one is not in the release notes. It falls out of dividing ByteDance's worked examples by their own token rates.

Their published five-second, 16:9, no-reference examples[1]:

Model480p720p
Seedance 2.5$0.514 ($0.103/s)$1.156 ($0.231/s)
Seedance 2.0$0.352 ($0.070/s)$0.756 ($0.151/s)

Divide each price by its token rate to recover the token count, then by 24/1024 to recover pixels. Seedance 2.5 lands at 9,607 tokens per second of 480p, roughly 854×480. Seedance 2.0 lands at about 10,057, a slightly taller frame.

720p resolves to 21,600 tokens per second on both, which is exactly 1280×720. That the same arithmetic produces a clean, verifiable number at 720p is what makes the 480p result trustworthy rather than a rounding artifact.

The consequence: at 720p, where the frame is unchanged, per-second cost rises by the same 53% as the token rate. At 480p, where the frame shrank about 4.5%, per-second cost rises only 46%. If you carry a per-second conversion factor from 2.0 to 2.5, it is wrong at 480p by roughly five percent, in your customer's favor if you are lucky and yours if you are not.

Your reference clip bills like generated video

input_video_seconds sits inside the same parenthesis as the output. A reference-to-video job pays for the source you uploaded at the identical rate as the frames the model invented.

Providers present this as a discounted per-second rate for reference mode, which reads as a bargain right up until you total it. Seedance 2.5's own published range makes the point without help: a five-second 720p generation costs $1.244 with a short reference and $4.838 with a 30-second one[1]. Same output, 3.9 times the invoice.

Seedance 2.5 also widened the input window to 30 seconds, where 2.0 caps at 15[1]. The ceiling on a single job roughly doubled alongside it.

There is a floor as well, and it is unpublished as a number. ByteDance's examples price two-second and four-second inputs identically, which implies a minimum around four seconds. The docs point at a spreadsheet calculator instead of stating it[1]. If you resell this, customers sending two-second clips cost you more than your arithmetic predicts.

What Seedance 2.0 costs per second today

For workloads that stay on 2.0, these are the current rates per billed second on reAPI[2][3]:

TierText or image inputWith reference video
Seedance 2.0 Mini 480p$0.038$0.024
Mini 720p$0.082$0.050
Fast Face 480p$0.062$0.036
Fast Face 720p$0.132$0.080
Seedance 2.0 Face 480p$0.07125$0.043125
Seedance 2.0 Face 720p$0.15375$0.09375
Seedance 2.0 Face 1080p$0.3825$0.2325
Seedance 2.0 Face 4K$0.780$0.480

Multiplied to a minute, that spans $1.44 to $46.80, a 32.5x range. Worth internalizing before you pick a default resolution for a product, because that decision compounds across every generation your users trigger.

Per-minute figures are a unit conversion rather than a purchasable thing, incidentally. A single generation caps at 15 seconds, so a minute of finished video is four clips minimum and realistically more once you discard the bad takes.

The practical move is to stop paying flagship rates while you are still hunting for the prompt. Mini at 480p is 19 cents for five seconds without source video. The same five seconds at 4K is $3.90. Parameters are aligned across the family, so iterating cheap and rendering expensive costs one model-id change. Twenty iterations at Mini plus one 4K render is $7.70; twenty iterations at 4K is $78.

FAQ

When does Seedance 2.5 launch?

It is already live. ByteDance introduced Seedance 2.5 on July 31, 2026, and reAPI exposes the production model as doubao-seedance-2.5-face.[5][6]

How much does Seedance 2.5 cost per second?

BytePlus's token examples work out to roughly $0.103 per second at 480p and $0.231 at 720p for a five-second clip with no reference video[1]. On reAPI, current no-source rates are $0.118589/s and $0.266824/s; source-video rates are $0.071153/s and $0.160094/s, applied to the documented billable clock.[6]

Does Seedance 2.5 support 1080p or 4K?

No. Only 480p and 720p appear on the current pricing page and reAPI production route[1][6].

Why is 4K cheaper per token than 480p?

Unexplained by the vendor, and it does not apply to Seedance 2.5 anyway, which publishes one rate covering both its resolutions. On 2.0, what matters is that the token count rises far faster than the rate falls, so 4K remains the most expensive output despite the lowest rate.

Is Seedance 2.0 being retired?

Nothing announced. Its rates and capabilities stand, and 1080p and 4K currently exist only there.

How do I estimate a job before running it?

Compute (input_seconds + output_seconds) × tokens_per_second from the pixel table, multiply by the rate, divide by a million. Then bill downstream on the returned usage.completion_tokens, since the formula is an estimate until the job finishes[1].

Does a failed generation still cost money?

On reAPI, no. Failed jobs refund automatically[3].

Budgeting for the switch

The honest summary is that BytePlus prices Seedance 2.5 above 2.0: 53% more per token and 46 to 53% more per second in its worked examples, with no 1080p or 4K production option. Whether that is worth paying depends on the value of 30-second output, expanded references, and editing in a specific workflow.

For a production estimate, rebuild the per-second conversion from the 2.5 pixel dimensions rather than inheriting the 2.0 factor, budget reference jobs on their full clock, and meter customers on returned usage rather than estimates. On reAPI, use the live per-second rate card and its documented minimum-billing formula.[6]

References

  1. BytePlus. ModelArk — model pricing (token rates, calculation formula, and per-video examples). Retrieved August 2026 from docs.byteplus.com/en/docs/ModelArk/1544106
  2. reAPI. Seedance 2.0 Face and Fast Face — model page and live pricing. Retrieved August 17, 2026 from reapi.ai/models/seedance-2-0
  3. reAPI. Seedance 2.0 Mini — model page and live pricing. Retrieved August 17, 2026 from reapi.ai/models/seedance-2-0-mini
  4. BytePlus. ModelArk — Seedance 2.0 API reference (resolution, duration and pixel dimensions). Retrieved August 2026 from docs.byteplus.com/en/docs/ModelArk/1520757
  5. ByteDance Seed. One-take Creation, Flexible Referencing: Introducing Seedance 2.5. Published July 31, 2026, retrieved August 2026 from seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5
  6. reAPI. doubao-seedance-2.5-face — live rates, model ID, resolutions, and source-video billing. Retrieved August 17, 2026 from reapi.ai/models/seedance-2-5 and reapi.ai/docs/seedance-2-5