# Models and effort

Source: https://modulify.ai/docs/build/models-and-effort

What the model picker offers, what effort changes, and how credits are charged.

The chat box has two pickers on its bottom row. The left one switches between Build and Plan. The right one, showing a small model icon, is the model picker. On a wide screen it also shows the current effort next to the model name.

## Models

Open the model picker and you get three entries.

| Name | Description in the menu | What it is |
|---|---|---|
| Auto | Adapts to your request | Modulify reads your request and picks Pro or Max for you |
| Pro | Sharp all-rounder | Claude Sonnet 5 |
| Max | The mastermind | Claude Opus 5 |

New projects start on **Max** with effort set to **Auto**.

Auto has no effort submenu because it chooses the effort as well. Pro and Max each open a submenu with the full effort list.

## Effort

Effort is how much reasoning the model spends before and while it works. More effort means slower answers and more credits, and better results on hard work.

| Name | Description in the menu |
|---|---|
| Auto | Picks the right effort |
| Quick | Basic, fastest, least spend |
| Brief | Light, fast, low spend |
| Steady | Solid, balanced, mid spend |
| Deep | Better, slower, high spend |
| Full | Best, slowest, top spend |

All six are available on both Pro and Max.

If you switch model while an effort is set, the effort carries across to the closest supported level rather than resetting.

## How Auto decides

When the model is Auto, the effort is Auto, or both, Modulify classifies your request into one of three tiers before the work starts. It reads the last six turns of the conversation, with the most recent one at full length, so a short follow-up like "now make it responsive" is still judged on the work it implies rather than on its word count.

| Tier | What it is for | Result |
|---|---|---|
| quick | One small change to something that exists, or a plain question | Pro at Quick |
| standard | Typical single-page building | Pro at Brief |
| complex | Full pages, real logic, data, integrations, debugging | Max at Deep |

The classifier is told to pick the higher tier whenever a request could fit two, because under-powering real work costs more than over-powering it. If the classification does not return in four seconds, the standard tier is used.

Auto only fills in what you left on Auto. Pin the model to Max and leave effort on Auto, and you get Max at whatever effort the tier suggests. Pin both and no classification runs at all.

### The first build is an exception

The very first generation of a brand new site, when the model is set to Auto, skips the classifier and runs at **Max with Full effort**. The first build sets the structure everything else is edited on top of, so it is never run cheaply.

The same two pickers sit on the dashboard prompt box, so this is a choice you can make before the project exists.

## What a turn costs

There is no flat price per message. Chat is metered on the tokens a turn actually uses, so a one-line copy change costs a fraction of what a full page build costs, on the same model.

Two things move the number:

- **The model.** Max costs two and a half times as much per token as Pro, both for what it reads and for what it writes.
- **The effort.** Higher effort produces more reasoning tokens, so the same request costs more at Full than at Quick.

Reading files, browsing, publishing and previewing are not charged. Only the generation itself is.

### Seeing the real number

Every finished assistant message records what it actually cost. Hover the message, open the three dots menu below it, and hover **Info**. It lists **Model**, **Thinking** (the effort that ran), **Duration** and **Credits**.

That row is worth checking when you are unsure whether a setting is worth it. The number next to Credits is the real charge for that turn, not an estimate.

### Your plan changes the rate

The same work costs half as many credits on the Pro Plan as it does on Free, Starter or Enterprise. The token usage is identical, the credits it converts into are not.

### Images are priced separately

Generating or editing an image costs **1 credit per image**, on top of the turn that requested it. Searching stock photography is free. See [Images](/docs/build/images).

## Where the setting is kept

Model, effort and mode are saved on the project, not on your account. Reopen the project tomorrow and the picker is where you left it, and a teammate opening the same project sees the same setting.

A queued message keeps the model, effort and mode that were selected when you queued it, even if you change the pickers afterwards. The queue row shows a small badge when a queued message differs from what the pickers currently say.

## Running out

When the workspace balance hits zero, the credits bar above the chat box switches to an **Upgrade** or **Top up** button, and pressing send opens that flow instead of sending the message.

If credits run out while messages are already queued, the queue stops rather than dropping them, and the panel reads `Paused because the workspace is out of credits.`

## Next

- [Plan and Build modes](/docs/build/plan-and-build) covers the other picker.
- [Writing prompts](/docs/build/writing-prompts) matters more than the model you choose.
- [Message queue](/docs/build/message-queue) explains what happens to messages sent mid-run.