Markups on the price¶

The concept behind models and routing is explained in Models and routing.

Markups is what your organisation adds to the price it is charged for a kind of usage -- for itself, or for every organisation below it. It sits beside Models, open to every administrator of the organisation. The price a markup is added to is the deployment's own: Artific writes it, your organisation reads it on Models, and a markup is what your organisation puts on top of it.

Whoever the markup applies to sees only the marked-up price, never that a markup is in it. Your own subtree summary shows both sides -- see Usage -- and so does your key's usage summary of one organisation below yours (Usage through the API).

Who a markup applies to¶

Every row chooses one of two:

This organisation changes only what your own organisation's summaries, caps and model prices show.
Everything below it changes what every organisation below yours sees as its own price, with nothing on its bill saying a markup is in it.

When more than one organisation in a chain sets a markup, they compound: a grandparent's markup on everything below it, then a parent's, then an organisation's own markup on itself, each on top of what came before it.

Within one organisation, rows never add up: for each usage, the most specific row in force applies on its own -- a model or a tool beats its kind, a kind beats every kind -- and among equally specific rows the latest. So "every kind +50%" beside "images +80%" charges images +80% and everything else +50%. The row that applies replaces the wider one whole: an amount row on one kind leaves that kind's other amounts unmarked, even if a percentage row covers every kind. This holds for each of the two choices above separately: your rows for this organisation and your rows for everything below it are weighed apart.

What a markup is for¶

A row names a kind of usage and, where that kind is charged by one, a model or a tool -- the same subjects a price names: model tokens and embedding tokens, charged per million by model; images, charged per image by model; transcribed audio, charged per minute by model; tool invocations, charged per invocation by tool name; and searches, storage, crawled pages, parsed pages and sandbox time, charged per metered unit by the kind alone. Unlike a price, naming none of them is allowed and means something: leave the kind empty to mark up every kind, or name a kind but leave its model or tool empty to mark up every one of that kind.

The two forms a markup takes¶

A row is either:

a percentage added to the resolved price -- 20 is +20%, -15 is a 15% discount, -100 is free, or
an amount added to it, in the same shape a price's own amounts take. For model tokens those are input, output, Cache-served input and Cache-written input per million tokens, each optional; a percentage scales all four.

Never both on the same row. A discount is only for Everything below it: this organisation itself cannot discount its own figure, since the one who caps it must never be the one lowering what it pays.

Add a markup¶

Choose New markup at the top right. The drawer asks, each in a card of its own:

Applies to: This organisation or Everything below it. What your choice changes is written under it.
For what: the kind of usage this markup applies to, or Every kind. Where the kind is charged by a model or a tool, the drawer asks for that too -- leave it empty to apply the markup to every one of that kind.
Form: Percentage or Amount on top, and the fields underneath change with the choice.
From: the moment the markup starts to apply. It may lie in the past -- every period back to it is then costed again with the markup in it -- or in the future, which announces a markup ahead of the day it applies.

Choose Record markup to save it.

The list¶

Every markup is a line of its own. It starts with the form: a percentage up in blue, a discount in green, an amount on top in yellow. Where one line adds several amounts, it shows the first; hover over it for the rest. Then come what the markup applies to, to whom and from when. On a phone each markup is a card.

What organisations below see¶

Choose a markup on Everything below it for one model that applies now, and a blue banner above the list shows three figures: what your organisation pays for that model, the markup, and what the organisations below see. The figure is the price per million input tokens, or per image for an image model. Beside the list, each organisation directly below yours that Core prices the model for is named with what it sees and why. Every figure is the price Core charges that organisation right now, with every markup in it -- the same price as on its own Models section. Where the organisations below do not all see the same figure, the banner says it differs and the card beside it shows each one.

The banner and the card are absent for a markup on this organisation itself, on every model, every kind or a tool, for a markup that does not apply yet or no longer applies, where your own organisation has no price for that model, and where no organisation directly below yours has one.

Change a markup¶

You do not. A markup is succeeded: choose Succeed in the row menu. The drawer opens with the same subject, scope and form; fill in the new value and a later start date. Everything from that moment on is costed with it, while the period before keeps what it was costed at.

Delete a markup¶

Choose Delete in the row menu. This is the correction path for a row that was typed wrong. Every period the row covered is costed again against what remains.

What it changes¶

A markup reaches everything that reads a price: your own usage summaries and caps, everything below you that the markup applies to, and the price your Models section shows. Nothing is retroactive beyond what From says -- a markup written today with today's date changes today onward, never what already happened.