Technical
What a Second of AI Video Costs: An October Price Check
ยท RenderBob team
A 4 October benchmark found a tenfold spread in billed price per second, and waits from 35 seconds to more than four minutes. Start the budget from the shot ratio.
Rate cards tell you little about the invoice. A benchmark published by Sandbase on 4 October 2026 compared billed prices and wait times for current video models on three product-clip prompts. It is a useful snapshot, and it is a small test.
The spread
Billed price per second of finished video ran from $0.03 (Veo 3.1 Lite, 720p, no audio) to $0.30 (Seedance 2.5, 720p with audio, after a 15 percent discount), a tenfold range. Median wait times ran from 35 seconds to 273 seconds. A four-second Veo 3.1 Lite clip cost $0.12 and finished in about half a minute. A five-second Seedance 2.5 clip cost $1.53 and took the longest. Wan 3.0 earned the best quality score in the test, 23 of 24, at $0.50 for a five-second 1080p clip at 30 frames per second, with a median wait of about 147 seconds.
Why rate cards mislead
Billing structures differ. Some providers charge per second, some by credits, and ByteDance bills Seedance 2.5 per token, so the cost moves with resolution and duration in ways a simple rate card will not predict. Audio is a separate line item at some providers. Other published comparisons show Veo 3.1 Standard around $0.40 per second, its Fast tier near $0.10 to $0.12, and Wan 3.0 listed between roughly $0.05 and $0.20 depending on resolution.
Turning this into a budget
The number that matters most is your shot ratio, not the rate. If a 60-second piece needs a 5-to-1 ratio of generated to used footage, you generate 300 seconds. At $0.10 per second that is about $30. At $0.30 it is about $90. Halving the ratio, by using draft tiers, tighter prompts, and good references, saves more than shopping for a cheaper model.
Worked example from the rates above, not a named production.
Quality per dollar
Tier the work: draft on a cheap or fast model, use a stronger model for shots that stay in the cut, and keep generic atmosphere on the lowest tier. In this test Wan 3.0 took the top quality score at $0.50 for five seconds, under the $1.53 Seedance clip.
Local versus cloud
Cloud pricing is predictable per second. Local generation has no per-clip fee and costs hardware, electricity, and time, and a local model may not match the frontier on quality. The comparison depends on volume. Occasional shots favour cloud. Steady high volume starts to favour local capacity.
Caveats
Three prompts is a thin sample, prices change monthly, and promotional discounts distort comparisons. Treat the numbers as a method for checking your own workload, not as a ranking.
More from the blog
- Nano Banana 2.1: 14 References, Three Thinking Levels, and Targeted Edits for Storyboards
Google's Nano Banana 2.1 accepts up to 14 reference images and adds a medium thinking level. Artlist also describes edits that change a named area instead of redrawing the frame.
- Odyssey-3 and Interactive World Models: What They Are, and What They Are Not Yet Good For in Production
Odyssey opened a public research preview of Odyssey-3 on 8 October. You move through a generated scene, which is a different planning job from ordering a finished clip.