A $1.00 vendor list price becomes $1.31875 on Sume: the formula

A **$1.00 vendor list price costs $1.31875 on Sume**. The steps are fixed: list price, minus any customer discount, times 1.25 for the model price, then a 5.5% platform fee on that model price. The result is a 31.875% premium over list, not 25% and not 30.5%.

The four steps

The rule is written down in docs/operations/list-price-plus-25-platform-fee.md and implemented in applyCustomerTerms in packages/provider-pricing. Every SKU that has a provider list price is multiplied by 1.25. The platform fee then applies to the post-discount bill, so the fee is 5.5% of the 1.25x number, not of the list number.

The order matters because the two percentages compound. 1.25 x 1.055 = 1.31875.

• Start with the provider list price in exact USD micros.
• Subtract a customer discount if your org has custom terms (none by default).
• Multiply by 1.25 to get the model price.
• Add 5.5% of the model price as the platform fee. The sum is the charge.

One clip, end to end

Take a 10-second Gemini Omni Flash 1.1 clip at 720p. The vendor page lists $0.10 per second at 720p, so the list price is $1.00. Sume's model price is $1.25, the fee is $0.06875, and the charge is $1.31875.

The same arithmetic across the four Omni resolutions for a 10-second clip:

Where the formula stops applying

Not every price is list x 1.25. Fastlane and HumanPost keep their own prices. A few calls are free today, including social scrape, voice clone and hypit notes and align. An org can also hold explicit all-zero price-book rows, which remove the fee for that SKU.

E2B and the agent's model sandbox have no second meter at all. The platform fee is what pays for them.

Why two percentages instead of one

The 1.25 multiplier is the price of the model call. The 5.5% fee is the same fee that applies to LLM agent turns, so one rule covers media jobs and chat turns. If you budget with a single number, use **1.31875 x list** and you will land on the charge to the micro.

Budgeting with the multiplier

The practical use is a one-line estimator. Take the vendor's list price for the call, multiply by 1.31875, and you have the charge. For a clip priced per second, multiply the per-second list by the seconds first. For a token-priced call, multiply the token price by the tokens and then by the multiplier.

A rounded 1.32 is close enough for planning, and it errs high by under a tenth of a percent. Use the exact figure when you reconcile against a usage row.

What the multiplier does not include

The formula is the price of the call itself. It does not add a seat price. Plans such as Pro at $40 per month are billed separately and set access and concurrency, not unit rates. It also does not add taxes, which depend on your billing details.

If you want to confirm the multiplier on your own account, run one small paid job and read the four price_book_* fields on its usage row. The model field divided by the list field should be 1.25, and the fee field divided by the model field should be 0.055.