Which model works on your ticket

Behind AdenAI there is not one single AI but several. They differ in strength, and they differ in what they cost.

So when you write a ticket you pick two things: which model works on it and at what strength. Both sit in the button below the text field, showing something like Opus · Ausgewogen (Balanced). Clicking it opens the list.

This article explains what that choice does and what it costs. How credits work in general is under Credits.

Your choice is a ceiling

The model on the ticket is not a price tag. It says how strong AdenAI is allowed to work on this task at most.

• Upwards it is sealed. Not a single step runs on a stronger model than the one you allowed. Choose Haiku and you never pay for Opus anywhere.
• Downwards there is room. Not every step needs the full strength. Where it buys nothing, AdenAI takes a smaller model.

So you get the strength of your model where it counts, and you do not pay for it where it is not needed. That is why the amount deducted is often lower than you expected.

What runs on your model and what does not

Step	Runs on
The plan: what gets built and in which order	your model
Design: layout, colours, type, everything you see	your model
The final look: reviewing the finished result and improving one thing	your model
The substructure: scaffolding, header and footer, forms	a middle model
The planning help, before the ticket goes off	a middle model
Short chores: questions back to you, legal pages, publishing	the smallest model

The bottom three rows never go beyond your choice. If you set Haiku, Haiku is what runs there too, and nothing else.

Decisions are always made on your model. Which colour, which typeface, which image: that is settled in the plan and in the design. Merely writing it down or downloading it afterwards is legwork, and it runs smaller.

A cheap step that is no good is more expensive than an expensive one. So there is a safety net: if a step on the smaller model fails, AdenAI repeats it once on yours. Working cheaper never means you get a worse result.

The planning help, before you send the ticket off

While your ticket is still in planning you can ask AdenAI two things without him building anything. AdenAI drüberschauen lassen (Have AdenAI take a look) reads your ticket and writes into the history what he is still missing in order to build. Daraus einen Auftrag machen (Turn this into a brief) condenses your ticket and your conversation into one text, which you then decide about.

Both run on the middle model, not on yours. A strong model is not needed for it: nothing is designed and nothing is built, a half-finished brief is read. Your choice still stays the ceiling: if you set Haiku, this runs on Haiku too. If you set Opus or Fable, it stays on the middle one.

It is billed like everything else:

• What is deducted is what that one call used. There is no fixed price per press of the button. Under Verbrauch (Usage) on the ticket it appears under its own name: Planungshilfe for the look, Planungsauftrag for the condensing.
• What does not work costs you nothing. If no suggestion comes of it, a short sentence appears in the history and nothing on your ticket changes.
• Pressing twice without changing anything costs nothing either. There already is a suggestion for this text. AdenAI shows you the same one again rather than paying for it twice. Write something new or reword something, and he takes a fresh look.

That makes this button the cheapest place in the whole process to improve your ticket. What a good ticket looks like is under How to write a good ticket.

The strength next to it

The model says who works. The strength says how long that someone thinks before writing anything down. You set it in the same menu, below the model list, where it is labelled Stärke (Strength). Change it later and your ticket history records it under that same name.

Strength	What for
Standard	No preference, the way AdenAI works on its own
Zügig (Brisk)	Gets straight to the task, the most frugal
Ausgewogen (Balanced)	Thinks where it matters, the normal case
Gründlich (Thorough)	Checks more before building. Takes longer and uses more
Sehr gründlich (Very thorough)	For tasks where the first attempt rarely suffices
Volle Sorgfalt (Full care)	Leaves nothing out, uses the most

Here too, your choice is a ceiling. A legal page does not think long about it, even with Volle Sorgfalt set.

Set the strength by the task, not by the wish. For "swap the phone number in the footer", Volle Sorgfalt buys you nothing but waiting time.

Why a higher number sits at the bottom of the ticket

On the ticket you will find the Verbrauch (Usage) panel. At the top is the number actually deducted from you. Below it comes a comparison headed Wenn ein einziges Modell alles gemacht hätte (If one single model had done everything), one card per model, and yours is marked.

Your card usually shows more than was deducted. That is not a mistake. The comparison works out what would have added up if every step had run on that one model, including the short chores. That is not what happened.

The two numbers answer two different questions:

• The upper one is measured. AdenAI measures every call individually. That sum is what leaves your balance, and nothing else.
• The lower ones are estimated. They take the same work and price it with the rates of each model. Whether another model would have needed longer for the same task, nobody knows. That is why each of those numbers carries a ~.

The card for your own model is an estimate as well, even though it looks like a measurement. It answers the question: "What would have added up if I had paid for this model on every small chore too?" The gap between that card and the number above it is what you saved.

The other cards are there for the opposite question: would a weaker model have cost you less, and how much less? You can decide by that on your next similar ticket.

What does this mean for your choice?

• A strong model on a small task still costs little. The amount follows the work, not the name in the menu. You are not buying a package, you pay for what was used.
• A weak model on a large task does not automatically come out cheap. If it lacks the power, it takes more attempts, and every attempt uses something.
• On the first ticket of a project the stronger model is especially worth it. That is where the foundation for everything else is laid. More on that under From idea to your first page.
• Which models you can pick depends on your plan. You still see the locked ones, with the plan next to them that unlocks each.

Changing it later

You can change model and strength at any time, including after sending the ticket off: pick it in the menu and click Übernehmen (Apply). A line then appears in the history recording what it was changed to.

A run that has already started does not switch models, though. It works to the end with whatever applied when it started. Your change takes effect from the next run on that ticket. The ticket header shows the new name immediately, and that is the one place where the display is ahead of the run.

If a number looks odd

Every entry is listed with date and reason under Plan & Credits, one line per ticket. If you find something there you cannot explain, write to us in support and give us the ticket number. We look at exactly the same figures you do.