Wan 2.5

What Wan 2.5 costs

Wan 2.5 is $0.05 per second of video through fal, which ties the cheapest published rate in this database – level with Veo 3.1 Lite at 720p and with the batch tier of Sora 2.

It is a flat rate: one price, no resolution ladder, no batch discount to reason about. A ten-second clip costs 50 cents and a full minute costs $3. For comparison, one minute on Sora 2 Pro at 1080p is $42.

Reading the aggregator rate correctly

This is the fal price for running the model, not the Alibaba console rate. Aggregators set their own margins and reprice independently of the model vendor, so the two figures can diverge in either direction and can move at different times.

For evaluation and moderate volume that is unimportant – the published rate is the rate you will pay. For a committed budget or a procurement process it is worth confirming against both the aggregator and the vendor, and worth noting that routing through an aggregator adds a dependency to your uptime and your data path that a first-party integration does not have.

What changes at five cents a second

There is a threshold in generative video below which the cost of generating stops being the constraint on a project, and $0.05 per second is under it for most work.

Consider a typical brief: twenty finished ten-second shots. At five attempts per keeper that is a hundred generations, a thousand seconds, $50. The review time to watch and judge a hundred clips is several hours of someone attention. The model is no longer the expensive part of the process – the person is.

That inversion should change how the work is run. When generation is expensive, the rational move is to prompt carefully and generate sparingly. When it is this cheap, the rational move is the opposite: generate widely, accept a low hit rate, and invest the effort in reviewing quickly and in building the selection step rather than in perfecting each prompt beforehand. Teams that carry cautious habits over from expensive models leave most of the value of a cheap one unused.

The retry maths, honestly stated

Cheap does not mean the multiplier disappears. Three to ten generations per usable shot is normal across every model on this site, so an accepted ten-second shot at five attempts costs $2.50 rather than 50 cents, and a ninety-second sequence built from nine such shots runs about $22.

What a $0.05 rate buys is tolerance for a poor hit rate. At ten attempts per keeper an accepted ten-second shot still costs $5 – less than a single first-attempt generation on Sora 2 Pro at 1080p, which is $7. A model that needs three times as many attempts can still be the cheaper choice by a wide margin when its rate is low enough, which is why per-attempt price and per-keeper price are worth calculating separately.

Where the cheap tier stops being the answer

Two things reliably push a project off the $0.05 band.

The first is resolution. If the deliverable needs 4K, Veo 3.1 is the only model here that generates it, at $0.30 per second on Fast – six times this rate, and unavoidable if upscaling is not acceptable.

The second is acceptance rate. If a cheaper model needs so many attempts that its per-keeper cost overtakes a dearer one, the rate card is misleading you. That crossover is real and it is workload-specific, which is why it is worth measuring on your own prompts rather than assuming. Generate the same ten prompts on Wan 2.5, Kling 2.5 Turbo Pro and Sora 2, count how many attempts each needs to reach something usable, and multiply by the rate. At these prices that experiment costs a few dollars and answers a question no pricing page can.