# Set up tiers and limits

> Choose which models serve each kind of work, in what order, and how much they may spend.

A tier is an ordered list of models with two spending limits. Plugins don't name models; their roles name a tier. Changing a tier changes every role that runs on it.

## The built-in tiers

| Tier | For | Starts at, per call | Starts at, per month |
|---|---|---|---|
| **Quick** | Bulk and simple work | $0.01 | $5 |
| **Balanced** | Most writing and reasoning | $0.05 | $10 |
| **Careful** | Hard judgement, final checks | $0.50 | $20 |
| **Frontier** | The hardest problems | $2.00 | $20 |
| **Research** | Needs the web | $0.05 | $5 |
| **Embedding** | Vectors for search | $0.01 | $5 |
| **Decision** | Typed yes/no, choice, score | $0.005 | $2 |

The starting limits are yours to change. The built-in tiers can't be deleted or renamed.

## Fill every tier at once

1. Open **Settings → Model Hub → Tiers**.
2. Under **Preset**, choose one and select **Apply**.

| Preset | What it sets |
|---|---|
| **Value** | Cheap, capable models at the bottom and Claude at the top, through OpenRouter. |
| **Claude** | Haiku, Sonnet, Opus and Fable, through OpenRouter first and Anthropic directly as the fallback. |
| **Local** | Models on your own computer. |

Applying a preset replaces the models and limits of all seven built-in tiers. Tiers of your own are left alone.

## Change one tier

1. On the **Tiers** tab, select the tier's row. Its dialog opens.
2. Under **Models**, the first model is the **Primary**. The rest are **Fallback 1**, **Fallback 2** and so on.
3. Use **Add a model** or **Add a fallback** to pick another. Use the arrows to reorder, and the remove button to take one out.
4. Under **Limits**, set **Per call** and **Per month** in dollars. Leave a field empty for no limit.
5. Select **Done**.

**Used by** at the bottom lists the roles that run on this tier, by plugin.

## How fallback works

Models are tried in order. When a call fails, or a model would go over the limit, the next model on your list is used. Model Hub never switches to a model that is not on the list. Every attempt is recorded, and the chat panel and the usage view show when an answer came from a fallback.

## How limits work

- **Per call** is checked before a call is sent, against the most the call could cost: the prompt plus the full length the reply is allowed. A call over the limit is refused and nothing is sent.
- **Per month** is this calendar month's spend on the tier. When it is used up, calls on the tier are refused until next month or until you raise it. The tier's dialog shows how much is spent.
- Local models cost nothing per call, so limits don't stop them.
- A role can have limits of its own. Limits stack: a call must fit the role's and the tier's.

## Per-model options

Select the sliders button on a model's row (**Routing and reasoning**) to set:

- **Reasoning**, **Output token cap** and **Context length**.
- For OpenRouter models: **Providers to try first**, **Providers never to use**, a **Price cap** per million tokens, **Only providers that keep no data**, **Let OpenRouter use other providers**, and **OpenRouter fallback models**.

## Make a tier of your own

1. On the **Tiers** tab, select **New tier…**.
2. Give it a **Name** (unique, at most 40 characters).
3. Add at least one model, set its limits and select **Create tier**.

A plugin can also ask to add a tier. The dialog then names the plugin, and nothing is added unless you select **Add tier**.

Open one of your own tiers to **Rename…** or **Delete…** it. A renamed tier keeps its roles, its limits and this month's spend. Deleting a tier sends each role you mapped to it back to its plugin's default.

## Related

- [Map a plugin's roles](roles.md)
- [See what you spent](usage-and-costs.md)