What is the Sume agent fee? The default 5.5%, explained

The Sume agent fee is a charge added on top of a request's model amount: usage is billed at each model's published rate plus a 5.5% agent fee by default, and a workspace price book can set another rate. Generation submit responses report it as its own line, agent_fee_amount_usd_micros, next to the model amount it was charged on.

Where the fee shows up and how to read it come from the Sume API reference, the OpenAPI schema that the API reference docs point to, read on 2026-09-27. The default rate is read from Sume's pricing code. For plans and the wallet, see How Sume pricing works.

Where does the agent fee show up?

Two places in the API schema name it. Generation submit responses in Sume's job envelope, such as those from POST /v1/avatar-1.0/talking-video or POST /v1/timeline-1.0/render, carry a usage estimate for that request. GET /v1/catalog can carry an applied estimate next to each list price, split the same way. Sume API catalog covers the rest of that endpoint.

How do I read the fee on one request?

Read it as a sum: billable_amount_usd_micros equals the model amount plus the fee. At the default rate, a model amount of $2.00 (2,000,000 micros) carries a fee of $0.11 (110,000 micros), so the billable amount is $2.11 (2,110,000 micros). That example amount is illustrative, not a Sume price.

• Micros are integers kept for exact storage: 1,000,000 micros is $1.00. Quote billable_amount_usd, which is always billable_amount_usd_micros / 1_000_000.
• billable_amount_usd_cents is ceil(micros / 10_000) cents for wallet rounding. It is not dollars.
• price_book holds the terms behind the amount: the workspace price book when one exists, otherwise the product default Agent Fee. Its agent_fee_bps states the rate in basis points, so the default reads 550.
• In the catalog, the applied total is model plus fee at micros precision, and each cents line is rounded up.

Is the fee refunded when a job fails?

The submit-time reservation stores the estimated Sume billable amount, which is model plus fee. Sume captures it on completion and refunds failed or pre-generation canceled jobs before capture. Do failed AI video generations cost credits? walks through reserve, capture, and refund.

Where is the fee not broken out?

Some responses report one total rather than a split:

• A Format run receipt's usage lists billable_amount_usd_micros, debited_usd_micros, holds, refunds, and the cap, with no fee field. What one Sume API run cost explains those totals.
• A /v1/videos job reports usage.cost when you poll it, which the docs define as the Sume billable amount.
• A completed POST /v1/images response reports usage.cost as the billed USD amount. A generation that outlasts the route's 30-second wait returns the job envelope instead.
• API pricing lists each capability's published rate; the agent fee is charged on top of those rates.