> Documentation index: [/llms.txt](/llms.txt). Every page as one file: [/llms-full.txt](/llms-full.txt).

# Choose a model

Choose a model [#choose-a-model]

The model picker in the composer decides which model writes the next reply; each reply names the model that wrote it.

<img alt="The Models list open from the composer, with Auto marked Recommended at the top, its description beside it, and a lock on Claude Haiku 5.5" src="__img0" />

The model picker [#the-model-picker]

* Select the model name next to the send button to open **Models**.
* **Auto** comes first, marked **Recommended**. The other models follow in alphabetical order, each with its provider's icon.
* Point at a model to read its description and, where available, its **Intelligence** and **Speed** ratings.
* The model you pick is used for your next messages until you change it; **New Chat** resets it. Choosing an agent switches to that agent's model.
* The label above each reply shows which model wrote it.

Auto [#auto]

Auto picks a model for every message you send, based on how demanding the request is.

| Rule           | Detail                                                                                                                                                          |
| -------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| What it weighs | Each message is sorted into one of three levels: fast, balanced or reasoning. A model is then picked for that level.                                            |
| Within a chat  | The level only goes up. Once a chat needs a stronger model, later messages in it keep at least that level.                                                      |
| Which models   | Only models your plan, your workspace and your organization allow. If no model fits the level, Auto takes the nearest lower level, then the nearest higher one. |
| Attachments    | When you attach images or PDFs, Auto prefers a model that can look at them directly.                                                                            |
| The label      | Replies show &#x2A;*Auto ·** and the model that answered.                                                                                                       |

<img alt="A reply labelled Auto · Gemini 3.8 Flash" src="__img1" />

Auto is not offered when your workspace allows no model that Auto can route to.

Locked and missing models [#locked-and-missing-models]

| What you see             | Why                                                                        | What to do                                                                                                                                                                   |
| ------------------------ | -------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| A lock next to the model | Your plan does not include it.                                             | Select it. **See plans** offers an upgrade; on a deployment without self-serve billing, or in an organization, the window offers **Contact Alomana**.                        |
| The model is not listed  | Your workspace's model list leaves it out, or your organization blocks it. | Ask your workspace or organization admin. See [Workspace models](/docs/platform/admin/workspace/models) and [Organization models](/docs/platform/admin/organization/models). |

<img alt="The Model not available on your plan dialog with Close and Contact Alomana" src="__img2" />

If the model you picked gets locked later, for example when your credits run out, the composer moves to a model you can still use. See [Usage and limits](/docs/platform/account/usage-and-limits).

An agent's own model [#an-agents-own-model]

An agent can have its own model, set by whoever built it.

* Picking the agent in the composer also picks its model.
* If your plan does not include that model, the composer uses the deployment's default model. If that one is locked too, it uses the least expensive model you can use.
* If your organization blocks the agent's model, the agent runs on the default model instead, without a message. The label above the reply shows the model that actually ran.
* A model you pick yourself after choosing the agent overrides the agent's model for your messages.

To set an agent's model, see [Configure an agent](/docs/platform/agents/configure).