Skip to main content
Settings → Models controls which models the agent runs on and how new chats start. The settings are personal: they apply to your own chats in the current workspace.
Agent settings Models section with Default Model set to auto, Small Model set to Use default model, Subagent Model set to Inherit from agent with its effort Not Set (Use Provider Default), Default Reasoning Effort Not Set (Use Provider Default) and Default Approval Mode Not Set (Use Last Choice)

Settings → Models

Open the Models settings

1

Open the settings

In AI Studio, click Settings in the sidebar. The settings dialog opens on Providers.
2

Go to Models

Select Models under Agent settings. On a narrow screen, pick it from the section menu at the top of the dialog.
You need at least one connected provider. Without one, the section shows “Connect an LLM provider to choose the models the agent uses.” with a Connect a provider button; see Connect a model. Changing these settings needs the agent.write permission.

Settings at a glance

Each change is saved as soon as you pick it. A model change confirms with Models updated; a reasoning effort or approval mode change confirms with Defaults updated.

Pick a model

The Default Model, Small Model and Subagent Model rows share the same picker:
1

Open the picker

Click the current value on the right of the row. The models of every connection load (Loading models…) and are grouped under the connection’s provider and name, for example OpenAI · My Work Key.
2

Find the model

Type in Search models… to filter by model ID or name.
3

Pick it

Click the model. A check mark shows the current choice. For Small Model and Subagent Model, the first entry (Use default model or Inherit from agent) clears the setting.

Default Model

The primary model for new conversations: the model a new chat runs on, and the model the composer’s picker starts on.
  • Options: any model of any connected provider.
  • Default: your first connection becomes the default when you connect it. If no default is set, or the default connection was disconnected, the oldest remaining connection is used with its own model, and the row reads Not set (uses the first connection).
  • Also used by: AI analysis of a vulnerability and AI tag suggestions on services.
The Default Model and the default connection in Settings → Providers are the same setting. Picking a model here also makes its connection the default and sets that connection’s model, so the composer shows it. In the other direction, Set as Default in Providers, or picking another model in the composer, updates this row. See Connect a model. The Default Model cannot be cleared, only changed.

Small Model

A lightweight model intended for title generation, conversation summaries and reports.
  • Options: Use default model, or any model of your connections.
  • Default: Use default model.
In the current release the setting is saved but not yet applied: titles and conversation summaries are generated with the conversation’s own model.

Subagent Model

The model, and optionally the reasoning effort, that subagents run with.
  • Options: Inherit from agent, or any model of your connections.
  • Default: Inherit from agent: subagents run on the same connection as the conversation that dispatched them, with that connection’s model.
When a subagent starts, OASM picks its model in this order:
  1. The agent’s own Override Model, if the agent being dispatched has one (see Agents).
  2. The Subagent Model set here.
  3. The conversation’s connection and its model.
  4. Your Default Model.
The model is resolved when the subagent starts, so a change here applies to subagents dispatched afterwards, including in running conversations.

Subagent reasoning effort

The second control in the row sets the reasoning effort for the Subagent Model: Not Set (Use Provider Default), Low, Medium or High. It is disabled while the row is on Inherit from agent.
  • Picking a different subagent model keeps the effort you set.
  • When the effort is not set, subagents use your Default Reasoning Effort, then the provider default.
  • If the Subagent Model is also the chat model of its connection and you picked an effort for it in the composer, the composer’s effort applies.

Default Reasoning Effort

The effort a new chat starts with when you leave the composer’s effort picker on Default. It also applies wherever no other effort is set, such as subagents without their own effort.
  • Options: Not Set (Use Provider Default), None, Minimal, Low, Medium, High, Extra high, Max.
  • Default: Not Set (Use Provider Default): the model runs at its provider’s default effort.
Models support different levels. A model that does not offer the level you set runs at its provider default instead. An effort picked in the composer for a connection always takes precedence over this setting; see Connect a model.

Default Approval Mode

How commands run when a new chat starts. Manual asks before each new command; Auto runs them all. See Plans and approvals.
  • Options: Not Set (Use Last Choice), Manual, Auto.
  • Default: Not Set (Use Last Choice): a new chat starts with the approval mode you last chose in the composer in this browser, or Manual if you never chose one.
The setting applies only when a chat starts. You can still switch the approval mode in the composer, and approving a plan with Yes, run it automatically or Yes, but approve each command switches the chat’s mode. Existing chats keep their own mode.
Setting Auto as the default means every new chat runs commands without asking. Built-in safety checks and plan approval still apply, but individual commands are not reviewed.

Change a setting

1

Open Settings → Models

In AI Studio, click Settings, then Models.
2

Pick the new value

Click the control on the right of the row and pick a model, effort or mode.
3

Check the confirmation

Models updated or Defaults updated confirms the save. The new value applies to the next chat you start, and the next subagent dispatched.
If the section shows “Couldn’t load the model settings.”, click Retry. Editing is disabled until the settings load, so a save never overwrites them with empty values.

Connect a model

Connect providers and switch model and effort in a chat

Agents

Give a custom agent its own override model

Plans and approvals

How Manual and Auto approval work

Usage

See the tokens and cost of each model