Sume vs fal direct: when the 31.875% premium is worth paying

Direct from fal you pay the vendor's list price. On Sume you pay list x 1.25 plus a 5.5% fee, which is **31.875% more**. For 500 ten-second Gemini Omni Flash 1.1 clips at 720p, that is $500 direct against $659.375 on Sume. Direct wins on price alone. The question is what the gap buys.

The gap in numbers

fal's page lists Omni Flash 1.1 at $0.10 per second at 720p. Ten seconds is $1.00 list.

When direct is the better buy

If you run one vendor's model at steady volume, have an account with that vendor and only need that vendor's output, go direct. You keep the full discount and avoid the multiplier. There is no argument against that.

When one wallet is worth the premium

The premium buys a single balance across image, video, speech, music and agent turns. It also buys a single usage ledger, a single API key, and the same error shape (a 402 insufficient_credits when the balance is short) across models.

Image billing is all-or-nothing, and failed or cancelled image generations are not charged. You can read GET /v1/balance and GET /v1/usage from one place for every model. A multi-vendor stack makes you build that yourself.

A rule of thumb

Count the vendors you would otherwise integrate. If it is one, compare 1.31875 x list against your own engineering time. If it is five, the premium is the price of not building four more billing integrations.

What stays the same

The premium does not change quality. The model is the same model. A clip generated through Sume and the same clip generated direct are not different products, so the comparison is only about cost and operations.

It also does not change launch promos on a vendor's own page. A vendor's temporary discount applies there, while Sume budgets from the ongoing list price. If a promo is big enough, direct will look even cheaper while it runs.

A quick checklist

Go direct if you use one vendor, already have an account, and have the engineering time to read that vendor's errors and billing.

Stay on Sume if you use several modalities, want one ledger to read with GET /v1/usage, or run an agent that needs to call media tools on one balance.