# Model classes

A model class is a tier of model capability: **Quick**, **Standard**, **Top** or **Ultra**. Each
job an agent does runs on one of them, and each tier is an ordered list of models, best first.
Productmind keeps a list for every tier. On **Model classes** you can see those lists and, if you
manage the workspace, replace any of them with your own, so your coding sessions and AI agents run
on the models you prefer. You can also lower how much reasoning effort your coding sessions use
with each model in a tier.

<Availability plan="Depends on your plan (Agent customization). If yours doesn't include it, the Model classes tab shows a lock and names the plan that does." permission="Agents access on the Roles tab, under Administrative access, to see the tab. Changing or resetting a list or an effort level: an Owner or Admin of your organization, the workspace's owner, or one of its admins." />

## Key concepts

**Tier.** How capable a model a job runs on. You choose the tier for each job on
[Templates](/user-guide/engineering/agents/templates#key-concepts):

| Tier | Means |
| --- | --- |
| **Quick** | The fastest and cheapest. Suits light, mechanical passes. |
| **Standard** | Balanced cost and capability. |
| **Top** | The strongest generally available model. The default for most jobs. |
| **Ultra** | The most capable model Productmind can reach. |

**Provider.** Who makes the models. Each tier has a list for each provider:
- **Anthropic**, Anthropic's Claude models;
- **OpenAI**, OpenAI's models.

Which list a run uses depends on the model provider it runs on. Each provider's section says which
AI agent sessions use it: Anthropic's list is for sessions running on Claude Code, and OpenAI's for
sessions running on Codex.

**The list.** The models of one tier on one provider, in order. The first model is tried first.
If it can't be used, the next one is, and so on down the list. The last model is the final
fallback.

**Platform default and Customized.** Every list starts as Productmind's own, the platform default,
and follows Productmind's changes to that tier automatically. Once you change it, it shows
**Customized**: your list replaces Productmind's, and no longer follows its changes. You can reset
it to the platform default at any time.

**No longer offered.** A model in your list that Productmind no longer offers. It keeps its place
in your list, marked **No longer offered**, and is skipped until you remove it or Productmind
offers it again.

**Effort level.** The most reasoning effort coding sessions use with a model in a tier, shown as
**Up to** a level, such as **Up to High**. The levels, lowest first, are **Low**, **Medium**,
**High**, **Extra high** and **Maximum**. Productmind sets a limit for each model in each tier,
which is its default. You can lower it for your workspace, to make sessions faster and use fewer
AI credits. A model Productmind sets no limit for can go up to **Maximum**.

## The Model classes tab at a glance

In the sidebar, under **Engineering**, choose **Agents**, then **Model classes**.

**The list of tiers** shows **Quick**, **Standard**, **Top** and **Ultra**, each with a one-line
description. A tier you've changed for any provider shows **Customized**, and a tier with models
Productmind no longer offers shows how many, such as **1 no longer offered**. Otherwise the tier
shows how many models it has. Choose a tier to open it beside the list.

**An open tier** shows its name and description, then one section per provider. Each section
shows:
- the provider's name, and **Customized** if you've changed its list;
- which AI agent sessions use the provider's list;
- the models, in order, each with its name, its model ID and its effort level, such as **Up to
  Extra high**;
- under a customized list, Productmind's own list for comparison, and who changed yours last.

The page lists only models you can choose. Productmind's lists can hold entries that aren't
models you can choose: the page doesn't show them, and they keep their place while you use
Productmind's list. Your own list never includes them.

If you can change the lists, each section also has **Add model**, a customized list has **Reset to
platform default**, and each model has an effort level you can change, in a column after its name,
**Move up**, **Move down** and **Remove** buttons, and a drag handle on the right of its row. The
header has **Reset to default settings** when any of the tier's lists or effort levels is changed.
If you can't, the tier says that only the workspace's admins can change model classes,
and each model's effort level shows as a label, such as **Up to Extra high**.

## Changing a tier's models

Each change saves straight away. The first change to a list that is still the platform default
saves your own copy of the list as the page shows it, with your change applied. A model that
Productmind's list still runs but no longer offers for your own lists is left out of your copy, and
the confirmation says so.

### To add a model

1. In the sidebar, choose **Agents**, then **Model classes**.
2. Choose a tier.
3. In the provider's section, choose **Add model**.
4. Choose a model from the list. Only models Productmind offers on that provider are listed.
5. Choose **Add model**.

The model is added at the end of the list, and the section shows **Customized**.

### To change the order

1. Open the tier.
2. Drag a model by its handle, on the right of its row, to its new place in the list. Or point at
   the model and choose **Move up** or **Move down** to move it one place at a time.

The new order saves as soon as you drop or move the model. The first model is tried first.

### To remove a model

1. Open the tier.
2. Point at the model, and choose **Remove** (the **X** button).

The model is removed from the list. A list needs at least one model: to go back to Productmind's
list instead, reset it.

### To reset a list to the platform default

1. Open the tier.
2. In the provider's section, choose **Reset to platform default**.
3. Choose **Reset**.

The section no longer shows **Customized**, and follows Productmind's changes to the tier again.
The provider's effort levels in the tier go back to Productmind's too. To reset every provider of a
tier at once, lists and effort levels, choose **Reset to default settings** at the top of the tier
instead.

## Changing a model's effort level

An effort level is a limit for one model in one tier. It saves straight away, and it doesn't change
the list: a list that is still the platform default keeps following Productmind's changes.

### To change a model's effort level

1. In the sidebar, choose **Agents**, then **Model classes**.
2. Choose a tier.
3. Next to the model, choose its effort level, such as **Up to Extra high**.
4. Choose a level.

The model shows **Up to** the level you chose. The list offers Productmind's limit for the model in
that tier, marked **(Productmind's default)**, and every level below it. Coding sessions in this
tier use the model at that level at most, from their next run. The same model in another tier
keeps its own level. If you remove a model from the list, its effort level goes with it: added
again, it starts at Productmind's level.

### To go back to Productmind's effort level

1. Open the tier.
2. Next to the model, choose its effort level.
3. Choose the level marked **(Productmind's default)**.

The model runs at Productmind's limit again. **Reset to platform default** and **Reset to default
settings** also put every effort level they cover back to Productmind's.

## How model classes behave

- **Your lists and effort levels apply to this workspace only.** Coding sessions and AI agents in
  other workspaces are unaffected.
- **A change applies from the next run.** A coding session already running picks it up at its
  next turn.
- **A coding session moves down the list when a model is busy.** If a model is at capacity, or
  isn't available to the account the session runs on, the session continues the same conversation
  on the next model in the list. The session's timeline says which model took over, for example
  **Claude Opus 5.5 was at capacity; continuing on Claude Sonnet 5.5.** This happens on every plan,
  over Productmind's lists where you haven't customized them.
- **An AI agent uses the first model in the list that it can run.** Some agents can't use every
  model: an agent that must produce structured output skips a model that always reasons first.
  When no model in your list suits it, the agent uses Productmind's model for that tier. An agent
  doesn't move to another model in the middle of a reply.
- **A model Productmind no longer offers is skipped.** If every model in a customized list is no
  longer offered, that tier runs on Productmind's list for that provider until you change yours. A
  model that is skipped has no effort level to change.
- **An effort level is a limit, and it applies to coding sessions.** A coding session uses a model
  at the lower of the job's own effort, which you choose on
  [Templates](/user-guide/engineering/agents/templates#key-concepts), and the model's effort level
  in the tier. If the session moves down the list, the next model runs at its own level. AI agents
  keep their own reasoning settings.
- **You can lower an effort level, never raise it.** Productmind's limit is the highest level
  offered. If Productmind lowers its limit for a model below the level you chose, Productmind's
  applies.
- **A LiteLLM model key keeps its own models.** When your organization's
  [model provider](/user-guide/engineering/agents/model-provider) is a LiteLLM proxy, the proxy's
  model for each tier applies, not these lists.
- **If your plan stops including Agent customization**, your lists and effort levels are kept but
  not used: runs go back to Productmind's lists and limits. They apply again once your plan
  includes it.

## Permissions

| To | You need |
| --- | --- |
| See **Model classes** and every tier's lists | **Agents** access (under **Administrative access** on the **Roles** tab) |
| Add, reorder or remove models, change effort levels, and reset a list or a tier | An **Owner** or **Admin** of your organization, the workspace's owner, or one of its admins |

The **Owner**, **Admin**, **Power User** and **Developer** roles have **Agents** access. A Power
User or Developer who doesn't manage the workspace sees the lists and effort levels but can't
change them. Lists and effort levels can only be changed by a person signed in to Productmind, not
through the API or an AI assistant. See
[Roles and permissions](/user-guide/reference/roles-and-permissions).

## Limits

| Limit | Value |
| --- | --- |
| Models in one list | 1 to 10 |

## Troubleshooting

### Add model, Remove and Reset are missing

Only an Owner or Admin of your organization, the workspace's owner, or one of its admins can change
model classes, so you also see each model's effort level as a label rather than a list. Ask one of
them, or ask to be made an admin of the workspace on its **Permissions** tab.

### A model's effort level doesn't go higher

The list offers Productmind's limit for the model in that tier and the levels below it. A model
that only accepts up to a certain level can't run above it, so the list stops there.

### A model shows No longer offered

Productmind no longer offers that model, or no longer offers it on that provider, so your sessions
and agents skip it. Remove it, or keep it in case Productmind offers it again.

### The Model classes tab shows a lock

Your plan doesn't include Agent customization. Open the tab to see which plan does. See
[Plans and limits](/user-guide/reference/plans-and-limits).

## Related topics

- [Templates](/user-guide/engineering/agents/templates)
- [Model provider](/user-guide/engineering/agents/model-provider)
- [Sessions](/user-guide/sessions/overview)
- [Plans and limits](/user-guide/reference/plans-and-limits)
