Unit Cost NotesField notes on the Wan 3.0 video API — sourced, dated, and corrected in public.

Resolution is a 4x price multiplier, not a quality slider

Alibaba's list price for Wan 3.0 is per second of output, and it moves with resolution:

Resolution	List price per second
480P	$0.05
720P	$0.10
1080P	$0.20

480P to 1080P is 4x, and the default is 1080P. Those two facts together are the whole point of this note: the expensive option is the one you get by not choosing.

What that does to a thirty-second clip

The duration range is 2 to 30 seconds, so the spread across a single request is wide:

• 30 seconds at 1080P — $6.00
• 30 seconds at 480P — $1.50
• 2 seconds at 480P — $0.10

A team iterating on prompts at the default settings is spending sixty times the cost of the cheapest useful probe, per attempt. Not because they chose quality, but because nobody touched the dropdown.

The iteration cost is where this actually bites

The mistake is not "we shipped at the wrong resolution". Nobody ships 480P by accident; you look at it. The mistake is iterating at delivery resolution.

Almost everything you check on an early pass is resolution-independent: did it understand the subject, is the motion sane, is the pacing right, did it put a caption where you wanted one, does the audio line up. All of that reads perfectly at 480P. What does not read at 480P is fine texture and small-text legibility — and those are the last things you tune, not the first.

So the workflow that falls out of the pricing:

1. Prompt exploration at 480P, short. You are testing whether the model understood you. Two to five seconds is usually enough to know.
2. Timing and structure at 480P, full length. Pacing problems only show up at full duration, and duration is the linear term.
3. One 1080P pass at the end, when the prompt is settled.

That is a 4x saving on the majority of your requests, and it costs nothing in output quality because the final render is at full resolution either way.

Two things that do not change the bill

Worth stating because both are widely assumed to:

Turning the audio off does not make it cheaper. Speech, effects and music are generated in the same pass as the picture. The switch changes the output, not the price — which tells you something real about how the model works.

There is no 4K tier. There are three resolutions and 1080P is the ceiling. Pages currently ranking for this model claim a fourth; it is not in any primary source. If you are budgeting against a 4K line item, delete it.

Checking the numbers

The price ladder, the duration range, the three resolution tiers and the audio behaviour are all published parameters. They are collected with sources and check dates on the Wan 3.0 specification table — read 2026-09-07 — alongside the claims in circulation that do not survive a look at the documentation.

Disclosure: this site is written by the people who run a third-party browser interface to this model. We are not affiliated with Alibaba or Alibaba Cloud.