What AI video generation really costs per second in 2026 (13 models, open data)

Short answer: at 720p, AI video generation costs between about $0.06 and $0.63 per second of finished video in October 2026, depending on the model. A 10-second 720p clip costs $0.92 on Grok Imagine, $2.00 on Wan 3.0, $2.50 on Kling 4.0, $3.50 on Seedance 2.5 and $3.67 on Kling 2.6. VEO 3.1 only makes 8-second clips: $2.00 (Fast) or $5.00 (Quality).

Full disclosure first: we build Kreavo, a platform that runs these models for UGC-style video ads. The numbers below are the prices our own billing formula charges (checked on 29 October 2026), not the model vendors' own API rates. We publish them as an open dataset under CC BY 4.0 because we could not find one place that listed them per second, side by side, with the length and resolution options that actually exist.

Model pages love "from $X per video". That number hides two things: how long the video is and at what resolution. A $1 clip can be 4 seconds at 480p or 10 seconds at 720p. If you are budgeting a product, a pipeline or an ad test, divide by the seconds you get. Everything below does that.

Lowest price per second at 720p for the main versions of each model (for flat-priced versions, the longest clip gives the lowest rate). A few extra variants, such as Grok Imagine 1.5 or Vidu Q3 Turbo, are left out here to keep the table readable; the dataset has all of them and the script further down prints the full list:

Unless noted, these versions generate sound with the video. The spread from the cheapest image-to-video option to the most expensive is about 11x. If you need pure text-to-video, the spread is still about 7x (Grok Imagine at $0.092 vs VEO 3.1 Quality at $0.625).

At 4K, Gemini Omni Flash costs $0.33 to $0.58 per second ($2.31 for 4 s, $3.30 for 10 s) and Gemini Omni 1.1 costs $0.75 per second ($7.50 for its 10-second clip).

Going from 720p to 1080p adds 40 % on Seedance 2.5, 48 % on Kling 4.0 and doubles the price on Wan 3.0, while Wan 2.6 Turbo and Gemini Omni Flash charge the same at both. If the video ends up in a 9:16 feed on a phone, test whether your audience can tell before you pay for it.

And one that no price table shows: retries. A clip you throw away costs exactly as much as one you keep. Budget per usable clip, not per generation.

Render time only counts if enough real jobs back it. Our bar is at least 10 customer jobs from 3 or more customers in the window. In the 30 days to 22 October 2026, two versions passed it (request to finished file, queue included, all lengths and resolutions together):

That is roughly 14 and 22 seconds of waiting per second of video. We do not publish a number for the other versions until they have enough jobs. These figures are the dated 22 October snapshot; the JSON below is refreshed more often, so its render-time fields can show a newer window.

Every version in the index carries a price_grid with every length and resolution it supports. This prints every version available at 720p, sorted, including the variants left out of the table above (Node 18+, no dependencies):

// Cheapest price per second at 720p for every AI video model version in the index const res = await fetch ( " https://kreavo.ai/api/model-index.json " ); const index = await res. json (); const rows = []; for ( const model of index. models. filter (( m ) => m. kind === " video " )) { for ( const v of model. versions ) { const at720 = v. price_grid. rows. filter (([, resolution ]) => resolution === " 720p " ); if (! at720. length ) continue; const perSecond = Math. min (... at720. map (([ seconds,, usd ]) => usd / seconds )); rows. push ({ version: v. name, usd_per_s: + perSecond. toFixed ( 3 ), start_image: v. requires_image }); } } rows. sort (( a, b ) => a. usd_per_s - b. usd_per_s ); console. table ( rows ); Enter fullscreen mode Exit fullscreen mode Swap "720p" for "1080p" or "480p" to get the other tables. The data is CC BY 4.0: reuse it, just credit the source.

Say you need 100 ten-second 720p clips a month:

Same volume, same resolution, a 5x difference. Most teams end up mixing: a cheap model to find the hook, a more expensive one to render the winner.

Kreavo is an AI UGC video tool for ecommerce brands: you keep one AI influencer, add your product photo, and render the scene on whichever of these models fits the budget. The full table, the per-second grid and two CSV files live on the AI video generation cost index, with the snapshot date on every row.

If you spot a number that looks wrong, leave a comment with the model, length and resolution, and we will check it against the formula.