Choosing a model & thinking effort

Behind every request you send the Assistant is a model doing the work, and you get to choose which one — and how hard it thinks. That choice trades speed against quality against the credits a turn costs, so it's worth a minute to understand.

Where the choice lives#

Every model and effort setting sits in one place. In the Assistant composer, click the sliders icon to open the Assistant options flyout. It's organised into sections — Before building, Model, and After building — and the model choice lives in the middle.

Click the Model row to open the Choose a model picker. A small dot on the sliders icon tells you at a glance when an opt-in option is active, so you always know when you've stepped away from the defaults.

The Choose a model picker#

The picker is a three-pane modal, subtitled "Pick the model that builds your design. Each has a different cost and speed."

• Provider sidebar — a Recommended shortlist at the top, then each AI provider grouped below.
• Model list — every model with its name, a short description, and its credit cost.
• Detail pane — a fuller description of the selected model plus its thinking-effort controls.

Clicking a row selects that model immediately — there's no separate confirm step. Close the picker and your choice is in force for the next request.

Let Auto decide#

Not sure which to pick? Choose Auto (the sparkle entry at the top). It "automatically picks the best model for your task" — a light model for a quick tweak, a stronger one for a full layout. Auto is the right default whenever you'd rather describe the design than shop for a model.

What credits cost#

Each model carries a credit multiplier, shown beside a coins icon as, for example, 2x credits, in both the model list and the detail pane. It's a relative cost: a higher-capability model spends more credits per turn than a lighter one.

That multiplier stacks with the size of the job — a big, multi-step build costs more than a one-line edit at the same model. Credits come only with a subscription; if you're weighing what a session will cost, see Plans & AI credits.

Thinking effort#

For models that support it, the detail pane shows an Adjust Thinking Effort control with three levels:

The hint says it plainly: "Higher effort = deeper thinking, slower response." More effort means the model reasons longer before it acts, which tends to produce cleaner results on hard problems — and takes more time and credits to get there. Not every model exposes effort; when it isn't shown, that model runs at its own fixed setting.

Which to pick, and when#

There's no single right answer, but a few rules of thumb cover most cases:

• Quick, obvious edits — "make the title bigger", "swap this color" — a light, fast model on Low. You'll eyeball the result anyway.
• Full designs and tricky print layouts — a stronger model on Medium or High. The extra thinking pays off when spacing, hierarchy and alignment all have to land at once.
• When you're unsure — Auto, and let the Assistant match the model to the ask.

One more thing to weigh: the model you pick also gates attachments. Some models can't accept image or document references, so if a request depends on an attached reference, choose a model that supports it.

Where the choice applies#

Your selection is remembered and applied per request. The Model row in the options flyout summarises what's active as Model · Effort, so a glance confirms your current setup without reopening the picker.

The choice also reaches beyond the editor: the composer on the landing page offers its own model dropdown, and whatever you pick there carries into the new design you're about to start.

Next#

Now that the mechanics are set, spend your credits well — write prompts that work so each turn lands on the first try. And to keep an eye on the bill, review Plans & AI credits.