Gemini Omni Flash 1.1 at $17.50 per 1M tokens: dollars per second

Gemini Omni Flash 1.1 video output costs $17.50 per million tokens on Google's paid tier, and Google's pricing page states that works out to roughly $0.10 per second of 720p video at 5,792 tokens per second. Input (text, image, video or audio) is $1.50 per million tokens. The same model is listed at $0.10 per second at 720p by fal. On Sume you do not count tokens: the model is billed per output second by resolution, at provider list times 1.25, reserved when you submit.

Sources: Google's pricing page and fal's Omni Flash 1.1 page, both read 2026-10-01, and Sume's Video Generation docs.

How do the token price and the per-second price line up?

5,792 tokens per second at $17.50 per million is $0.1014 per second, which Google rounds to about $0.10. Google's page gives the token rate for 720p only, so this post does not convert 1080p or 4K; use the per-second numbers below for those.

What does a clip cost at each resolution?

fal publishes a per-second price for each of the four resolutions Omni Flash 1.1 offers. Multiplying by Sume's documented 1.25 house rate gives the amount Sume would reserve. Treat the right-hand column as arithmetic, not a quote: the live rate is in pricing_skus from GET /v1/videos/models.

Which number should you budget with?

If you call Google directly, the token bill follows the output you receive, and input tokens are extra whenever you send reference images, clips or long prompts. On Sume the reserve is known at submit: the job carries a fixed per-second price, and Sume's docs say failed jobs release or refund the reservation. A 3 to 10 second Omni clip at 720p therefore reserves between $0.375 and $1.25 under the arithmetic above.

The Sume docs also describe usage.cost on the poll response as the billable amount, so you can reconcile after each job instead of estimating.

Worked example: draft small, finish large

A common pattern is to test a shot cheaply and only pay for the high-resolution render once the framing is right. Under the arithmetic above, a 5-second draft at 360p reserves $0.1875 on Sume, while the same shot at 1080p reserves $0.9375. Three throwaway drafts and one final cost $0.5625 plus $0.9375, or $1.50 in total, which is the price of one 8-second 1080p clip.

Because both resolutions are the same model id, the only field that changes between draft and final is resolution. Keep the prompt, the first frame and the duration identical so the comparison is fair, and send a fresh Idempotency-Key for each render so a retry of the final does not collide with a draft.

Limits

Google's page may add tiers, batch discounts or free-tier terms that this post does not cover. The fal figures are a third party's list, not Google's, and Sume's rate can change if the provider list changes. Always read the catalog before you commit a budget.