How many Shorts episodes does $100 buy on Wan 3.0 and Omni Flash?

At Sume prices $100 buys 1,600 seconds of Wan 3.0 at 480p, 800 at 720p and 400 at 1080p. That is 26.7, 13.3 and 6.7 sixty-second episodes of generation only.

The table covers both Wan 3.0 and Gemini Omni Flash 1.1, from the same rule: provider list price times 1.25 per output second.

The table

Divide $100 by the per-second rate for each tier. Output seconds are what you pay for, so a take you throw away is spent the same way as a take you keep.

YouTube's Help page on Shorts series describes a series as a Shorts-only playlist with seasons and episodes, with each episode at most 180 seconds (read 2026-10-07).

Read it as an upper bound

The numbers assume no re-rolls, no captions, no renders and no audio jobs. Captions are $0.20 a clip, a Timeline render is $0.10 per started minute. A real season spends a share of its money on takes you do not use, so plan for the ratio of takes kept to takes made.

Pick the cheap tier for choosing and the dear tier for the final, as in the season cost post. Check the API pricing page for any price that has changed.

• Budget a re-roll rate before the season starts.
• Send resolution; Wan falls back to 1080p in the estimator.
• Set a per-run cap with the spend-cap post.

Turning the table into a plan

Pick the tier by purpose. A draft at Omni Flash 360p costs $0.0375 per second, so $100 buys over 2,600 seconds of drafts. The same $100 buys 400 seconds of Wan 3.0 at 1080p. Drafts and finals use the same ideas at very different prices.

A workable split is 70 percent of the money on drafts and 30 percent on finals. Then the table reads simply: $70 of drafts at 480p Wan is 1,120 seconds, and $30 of finals at 720p is 240 seconds, which is four 60-second episodes. Your own ratio will depend on how many drafts you throw away.

Round down. The estimator may reserve the higher figure for requests that omit a resolution, and a reservation is held against your balance until the job is settled.

Checks before you spend

Before you spend, confirm three things: the model is in the catalog for your account, the resolution you want is in its supported_resolutions, and the duration you plan is in supported_durations. A request outside those fields is rejected, and a rejected request is the cheapest mistake you can make.

Then run one 10-second take at the lowest tier, read it, and scale. The rest of the $100 is only worth spending on a look you have already approved.