What a second costs on fal

fal (fal.ai) runs other companies' video models and prints its own figure beside each one: $0.05 a second on Wan 2.5, $0.07 on Kling 2.5 Turbo Pro and $0.4 on Veo 3. One model on the same list, Ovi, is priced by the clip rather than by the second. As of 2026-09-22.

Inclusion rule. Tiers the vendor publishes a per-second rate for. A tier priced only as a monthly allowance is not converted and does not get a row. Order. As the vendor orders them on its own page.

1How many priced tiers each channel publishes

Inclusion rule. Channels that publish a per-second rate, or a credit rate together with a credit price. Tiers are counted as the pricing page itself breaks them out. Order. Alphabetical by channel name.

2The page names its unit, which almost none of them do

One sentence on this price list does something unusual: it names the meter itself, saying that the output unit differs from model to model and may be a second or a whole clip. Most pricing pages in this register leave the unit to be inferred from a column header, and inferring it wrongly is how a comparison ends up dividing a clip price by a duration nobody published.

Saying it out loud also makes the Ovi row legible. A figure of $0.2 against a clip is not a worse price than $0.05 against a second; it is a different measurement, and on a five-second clip it lands between the two cheapest rows here. This ledger keeps it in the tier column rather than converting it, because the duration that would make the conversion true is not printed next to the figure.

3An averaging note sits under the comparison, and it matters

Beneath the figures the page states that they rest on an estimated average video of five seconds at 720p, and that actual output varies by model, resolution and prompt complexity. That is a caveat about the comparison, not about the per-second rates, and the distinction is worth holding: the seconds are billed as seconds, while any cross-model figure the page offers is standing on an assumed clip.

For a production the note reads as a warning about resolution. A rate quoted against 720p is the rate for 720p, and this page does not break its video figures out by resolution the way the model owners' own pages do. Anyone budgeting 1080p here is budgeting against a figure the page did not attach to 1080p.

4The same models, resold, at different numbers

Three of the four rows here belong to models that appear elsewhere in this register under their makers' own prices. Wan at $0.05 a second matches what its maker's documentation prints for 480P; Veo 3 at $0.4 sits between two of Google's published Veo tiers. Nothing about that is irregular, and it is the reason a rate is recorded against the seller rather than against the model: the buyer pays this page, and the terms on it are this page's.

What the reseller row cannot tell you is which resolution or variant the figure covers. Where a model owner publishes a ladder and a reseller publishes one number, the number is a point on somebody else's ladder, and this page does not say which rung. No maximum clip length appears either, so that column stays empty here.

5Sources

Read from the pay-per-use prices published at fal.ai, checked 2026-09-22. How credits are converted to dollars is on the method page; every channel side by side is on the ledger. What fal publishes in the other columns: what the page meters. Families fal prices, set against every other seller of the same weights: Wan, Kling, Veo. Next along the ledger: Replicate, Novita AI.

• How the page names its own billing unit
  Video models are billed by output unit, per second or per video, depending on the modelfal, pay-per-use prices / recorded 2026-09-22
• Wan 2.5
  $0.05 per secondfal, pay-per-use prices / recorded 2026-09-22
• Kling 2.5 Turbo Pro
  $0.07 per secondfal, pay-per-use prices / recorded 2026-09-22
• Veo 3
  $0.4 per secondfal, pay-per-use prices / recorded 2026-09-22
• Ovi
  $0.2 per videothe model on the page priced by the clip insteadfal, pay-per-use prices / recorded 2026-09-22
• The assumption printed under the comparison
  Figures rest on an estimated average video of 5 seconds at 720p, and the page says actual output may vary by model, resolution and prompt complexityfal, pay-per-use prices / recorded 2026-09-22
• Longest single generation
  Not stated on the vendor's price page (as of 2026-09-22)absent from the page carrying the ratefal, pay-per-use prices / recorded 2026-09-22