> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dust.tt/llms.txt
> Use this file to discover all available pages before exploring further.

# Model access tiers

> Restrict which models and reasoning efforts your members can select, for the whole workspace, a group, or one member.

Members can pick the model an agent runs on, from the conversation input bar and from the agent builder. See [Auto models and the model picker](/docs/user-documentation/agents/model-selection).

As an admin, you can cap how far up that list each member can go. This is useful when you want everyday work on inexpensive models and want frontier models reserved for a small group.

## The three tiers

Every model, at every reasoning effort it supports, belongs to one tier:

| Tier               | What it contains                                                                                     |
| ------------------ | ---------------------------------------------------------------------------------------------------- |
| **Cost efficient** | Models that trade some performance for a lower price, while remaining usable for everyday Dust work. |
| **Balanced**       | Standard, everyday workhorse models with good quality for most tasks.                                |
| **Premium**        | Higher-cost combinations for demanding work that needs stronger capabilities.                        |

Tiers are cumulative. Granting **Up to Balanced** also grants Cost efficient. Granting **Up to Premium** grants everything, which is the default for every workspace.

<Note>
  A tier applies to a model **and** a reasoning effort, not to a model alone. The same model can sit in two tiers: for example a mid-size model at Light effort can be Cost efficient, at Medium effort Balanced, and at High effort Premium. A frontier model is Premium at every effort.
</Note>

## Set the workspace ceiling

1. Go to **Admin** > **Usage**.
2. In the **Models tier** section, set **Workspace access** to the highest tier your members may select.
3. The change applies to every member who has no group or personal override.

The default is **Up to Premium**, so nothing is restricted until you lower it.

## Override for a group

On the same **Usage** page, the groups table lets you set a tier per group. A group set to **Inherited from workspace** follows the workspace ceiling.

Group overrides are how you give a specific team access to Premium models while the rest of the workspace stays on Balanced.

## Override for a member

The members table on the **Usage** page lets you set a tier for one member. Leaving a member on **Inherit** makes them follow their groups, or the workspace if they have no group override.

## How the ceiling is resolved

Dust resolves one ceiling per member, in this order:

1. **Member override**, if set. It wins over everything else.
2. **Group overrides**, if the member belongs to at least one group with an override. When a member is in several such groups, the **highest** of those tiers applies.
3. **Workspace access**, otherwise.

## Let members run published agents above their tier

Restricting a member also restricts the agents they can run: if an agent is configured on a model above the member's ceiling, the run is blocked with **Model tier not enabled**.

That is often too strict, because a well-built shared agent may deliberately use a strong model. In the **Models tier** section, enable **Published agents** to let every member run published agents whatever the agent's model tier, while still capping what they can select themselves.

With **Published agents** enabled:

* Published (shared) agents run normally for everyone.
* Unpublished and personal agents remain restricted to the member's ceiling.
* Creating or editing an agent on a model above the member's ceiling remains blocked.

## What members see

* In the model picker, models and reasoning efforts above their ceiling are visible but locked, with a tooltip telling them to contact their administrator.
* When they try to run an agent above their ceiling, they get a **Model tier not enabled** message naming the agent.

## Audit trail

Changes to model tier settings are recorded in your [audit logs](/docs/user-documentation/admins/audit-logs/audit-logs), so you can see who changed a workspace, group, or member ceiling and when.

## Availability

Model access tiers are available on usage-based (credit) plans. On legacy plans, premium model and reasoning-effort combinations are not selectable through the model picker at all, and there is no tier setting to configure.

## FAQ

### Do I have to configure anything for members to use the latest models?

No. Every workspace starts at **Up to Premium**, which restricts nothing. Configure tiers only if you want to limit what members can select.

### Which models are Premium, and which are Cost efficient?

It depends on the model and the reasoning effort. You can get the full list by clicking on the (i) icon near model tier selection, in the Usage page. As a rule of thumb: small and mini models are Cost efficient, mid-size models are Cost efficient or Balanced at low effort and Premium at their highest effort, and frontier models are Premium at every effort. The tiers evolve as providers release new models and change prices.

### Does restricting tiers break agents that were already built on premium models?

The agents keep their configuration. What changes is who can run them: a member below the required tier is blocked unless the agent is published and you have enabled **Published agents**.

### What happens when a member belongs to two groups with different tiers?

The highest of the two applies. Group overrides grant access, they do not subtract it. To cap one specific person below their groups, set a member override.

### Can I restrict a single person rather than a group?

Yes. Set a tier directly on that member in the members table on the **Usage** page. A member override takes precedence over their groups and over the workspace ceiling.

### Does this reduce credit consumption?

It caps the most expensive combinations your members can reach on their own, which removes one source of unplanned spend. It is one lever among several. See [Optimize credit consumption](/docs/user-documentation/admins/usage-seats-and-credits/optimize-credit-consumption).

### Does restricting tiers also restrict the auto models?

No. Complex and Balanced tiers will follow the constraint and pick models accessible to the users.
