Seedance 2.5 is live — 30-second cinematic video with native audio & real-person references
Wan 3.0 API Pricing: 480P, 720P, and 1080P Costs
2026/08/23

Wan 3.0 API Pricing: 480P, 720P, and 1080P Costs

Compare Wan 3.0 official API prices with reAPI rates, then calculate 5-, 10-, and 30-second video costs at 480P, 720P, and 1080P.

Wan 3.0 costs $0.038, $0.076, or $0.151 per output second on reAPI at 480P, 720P, and 1080P. Alibaba's published rates for the same three resolution tiers are $0.05, $0.10, and $0.20 per second, so the live reAPI rate is about 24% lower across the table.[1][2]

Those rates apply to text, frame, multimodal reference, document, and web-page inputs. Resolution and output duration set the bill; turning generated audio on or off does not change it.

Wan 3.0 price per second

Prices checked August 23, 2026:

ResolutionAlibaba published pricereAPI pricereAPI saving
480P$0.050/s$0.038/s24%
720P$0.100/s$0.076/s24%
1080P$0.200/s$0.151/s24.5%

This is a gateway price comparison, not an Alibaba price cut. Alibaba's public rate and reAPI's public rate are separate offers for access to Wan 3.0.

What a 5-, 10-, or 30-second clip costs

The simple calculation is:

video cost = per-second rate × requested output duration

At the current reAPI rates:

Duration480P720P1080P
5 seconds$0.190$0.380$0.755
10 seconds$0.380$0.760$1.510
30 seconds$1.140$2.280$4.530

The equivalent Alibaba published-price totals are $0.25/$0.50/$1.00 for five seconds, $0.50/$1.00/$2.00 for ten seconds, and $1.50/$3.00/$6.00 for 30 seconds.

The maximum duration matters here. Wan 3.0 can generate 2–30 seconds in one request, so a 30-second result does not require paying for two 15-second jobs and then joining them.[3]

Two defaults can make the estimate wrong

The public API defaults to 1080P when resolution is omitted. It also accepts duration: -1, which lets the model choose the length. On reAPI, that unknown-length request is reserved at the 30-second ceiling because the output duration is not known at submission time.[1]

For a predictable bill, set both fields:

{
  "model": "wan3.0-video",
  "prompt": "A quiet train platform after rain, one continuous tracking shot",
  "resolution": "720P",
  "duration": 10,
  "audio": true
}

That request is $0.76 at the current rate. Omitting the resolution changes it to a $1.51 1080P request. Using duration: -1 reserves $4.53 at 1080P.

Reference media does not create a second rate card

Wan 3.0 accepts up to ten reference images, five video clips, five audio clips, or one supported document or public web page. The input type does not change the per-second price.[1]

There is still one duration constraint to budget around: when a reference video is present, its total duration plus the requested output must fit inside 30 seconds. A 12-second source clip leaves at most 18 seconds for the output.

This is simpler than APIs that bill both source and output duration. It does not mean every job has the same production cost, though. A difficult reference task may take more attempts, and retries are usually the largest difference between a spreadsheet estimate and the cost of an accepted shot.

Budget attempts, not just the final MP4

A rate card tells you the cost of one completed attempt. Production budgeting needs an acceptance rate.

Suppose a team needs ten approved 10-second clips at 720P. One pass across the shot list costs $7.60. If half of the attempts survive review, the team needs roughly twenty generations and the expected generation spend becomes $15.20. At a 25% acceptance rate it becomes about $30.40.

expected generation cost
  = number of approved clips ÷ acceptance rate
  × clip duration × per-second rate

That formula also shows when a high-resolution retry policy is wasteful. If the team explores at 480P and promotes only one attempt per shot to 1080P, ten shots with four 480P attempts and one 1080P final each cost:

40 drafts × 10s × $0.038 = $15.20
10 finals × 10s × $0.151 = $15.10
total                         $30.30

Running all fifty attempts at 1080P would be $75.50. The staged workflow saves $45.20, provided the lower tier exposes the failures the team cares about. Tiny typography and fine face detail may still need an earlier 1080P test.

Cost controls for an API product

Do not let users submit an unbounded combination and discover the total after polling. A useful product can calculate the maximum reserve before submission:

  • require an explicit resolution rather than accepting the 1080P default silently;
  • show rate × duration beside the generate button;
  • warn that duration: -1 reserves at 30 seconds;
  • cap retries or daily spend at the application level;
  • log the requested duration, resolution, task id, settled charge, and outcome;
  • separate prompt experiments from final-resolution queues.

For a batch pipeline, place the estimate on the batch, not only each item. A CSV containing 500 ten-second 1080P rows represents up to $755 of first-pass generation at the displayed rate. That is worth an approval step even when every individual row looks inexpensive.

How to compare a first-party and gateway price correctly

The model may be the same while account terms, minimum purchase, regional billing, rate limits, support, and API envelope differ. Price per second is one comparable field, not proof that two services are interchangeable.

Record the exact date, resolution, mode, and billing unit for both sides. Do not mix a promotional credit bundle with a cash rate, or a 720P figure with an adaptive/default resolution. And do not describe reAPI's lower access rate as though Alibaba changed its own list price. The honest statement is narrower: on August 23, 2026, reAPI's displayed per-second rates were about 24% below Alibaba's displayed international figures used in this comparison.

Which resolution should you start with?

  • 480P is the cheap prompt and motion check. A five-second attempt costs $0.19.
  • 720P is a sensible working tier when framing, action, and timing still need iteration.
  • 1080P is for the accepted prompt or work where final resolution matters immediately.

Starting every prompt at 1080P pays four times the 480P rate before the scene direction is settled. A practical loop is to test at 480P, rerun the best prompt at 720P, and reserve 1080P for the version likely to ship.

You can inspect the live rate card on the Wan 3.0 model page and the full request surface in the Wan 3.0 API documentation.

FAQ

Is Wan 3.0 billed by input or output length?

Ordinary requests are billed by requested output seconds. Reference video is constrained by the 30-second combined input-plus-output window, but the public reAPI rate is still multiplied by output duration.

Does Wan 3.0 audio cost extra?

No. The audio switch does not change the per-second rate.

Are failed Wan 3.0 requests charged?

The reAPI documentation states that failed jobs are refunded automatically.[1]

References

  1. reAPI, “Wan 3.0” model page and Wan 3.0 API documentation, accessed August 23, 2026.
  2. Alibaba Cloud Model Studio, “Wan 3.0 Video” release and pricing, accessed August 23, 2026.
  3. Alibaba Cloud, “Wan 3.0: 30-Second AI Video Generation from Any Input”, 2026.