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

# Choose a model

> Choose a Harness, model, and reasoning effort in the input box, learn how to switch models after a session starts, and understand Ultra and Ultracode.

A **Harness** is the framework that runs the Agent, such as Claude Code or Codex CLI. The **model** is the large language model the Harness calls. Before you start a session, choose both in the input box, then adjust the reasoning effort to match the difficulty of the task. For who provides the model and where usage is counted, see [Model sources](https://docs.aihubmix.com/en/cloud/models/model-sources.md).

## Choose a Harness and model

**1. Open the model picker**

On the new session page, click the model button in the bottom right of the input box. The button shows the current Harness and model name. "Subscription" after the model name means it comes from your subscriptions, and "BYOK" means it comes from an API key you added. Plus models have no label.

**2. Choose a Harness**

Click a Harness in the top part of the picker. When you change the Harness, the model resets to the **Default model**.

**3. Choose a model**

In the bottom part, switch between the **Subscriptions** and **BYOK** tabs and click a model. Plus models and your own API keys are both in the **BYOK** tab, with Plus models in the first group.

**4. Send**

Enter your task and send it. The session starts with this combination.

The Harnesses currently available are Claude Code, Codex CLI, OpenCode, Hermes Agent, DeepSeek, and Pi, as shown in the app.

A new session uses the Harness, model, and reasoning effort from the last session you created. The first time, it uses Claude Code with the default model. If the model you chose before is no longer available, it resets to the default model.

### Which models work with which Harness

The picker only lets you choose models that the current Harness can use:

* **Your subscriptions** pair by subscription type. For example, a Claude subscription works only with Claude Code, and a ChatGPT subscription works with Codex CLI.
* **Your own API keys** pair by API format: Claude Code needs an Anthropic-compatible API, and Codex CLI needs an OpenAI-compatible API.
* **Plus models** available vary by Harness.

Incompatible groups appear at the end of the list with the reason and cannot be selected. In a new session, clicking a model in one of these groups switches to a Harness that can use it and selects the model. Watch for the notice.

> [!NOTE]
>
> Without Plus, the picker opens as usual, but the Plus model group cannot be selected. Click **Get Plus** under the group to subscribe. See [Plus plan](https://docs.aihubmix.com/en/cloud/account/plus.md).

## Adjust reasoning effort

Reasoning effort controls how long the model thinks before it answers. Higher levels give more thorough answers but are slower and use more. Click the effort button next to the model button to choose. The button is hidden when the model does not support adjustment.

The effort menu comes in two styles: Claude models show a list of **Effort** levels, and GPT models show an **Effort** slider. Other models use the list in Claude Code and the slider in other Harnesses.

How to choose:

* When unsure, use the level marked **Recommended**.
* Use a low level for small, simple edits to get faster responses. Use **Medium** for most tasks.
* Use **High** for complex changes, and **Extra high** for cross-file refactors and hard problems.
* **Max** (in the list only) can use a large number of tokens and respond very slowly. Use it only for the hardest tasks.

The levels each model offers and which one is recommended depend on the combination of model, source, and Harness, as shown in the app. When you switch models, your level stays the same if the new model has it. Otherwise it resets to the new model's recommended level.

![Reasoning effort slider](https://mintcdn.com/aihubmix/ADcIS2RBGw_ogRXt/images/cloud/en/effort-menu.png?fit=max\&auto=format\&n=ADcIS2RBGw_ogRXt\&q=85\&s=c5857ffd7d39045edd7cd28909df3a2f)

Reasoning effort slider: drag to choose a level, with a description of what the current level suits below

### Ultra and Ultracode

Both make the Agent work on the task in parallel, use noticeably more, and suit truly difficult work.

* **Ultra**: The highest level for some models in Codex CLI, at the far right of the Effort slider, with multiple subagents working in parallel. Usage is about 2 to 3 times that of Extra high, and each task takes longer. You are asked to confirm when you choose it. Click **Turn on Ultra** to apply.
* **Ultracode**: The multi-agent parallel switch in Claude Code, which splits the task into several subtasks completed at the same time. It sits below the Effort level list and does not change the level. When it is on, the effort button shows something like "High · Ultracode". It requires all three of these: the Harness is Claude Code, the cloud server is ready, and the selected model supports Extra high. When any condition is not met, the switch is unavailable and shows the reason. Ultracode is off by default in new sessions.

## Switch models during a session

After the first message is sent, **the Harness is fixed**. To use a different Harness, start a new session. If the session belongs to a project, click **New task with another harness** in the model picker to start a new session in the same project with another Harness.

You can still change the model, source, and reasoning effort during a session. Changes apply from the next message, affect only this session, and do not change the defaults for new sessions.

**1. Wait for the Agent to finish the current task**

You cannot switch models while the Agent is working.

**2. Choose a new model**

Click the model button and choose another model. You can choose across sources.

**3. Confirm the switch**

If the current context fits in the new model, it switches directly. The full session history is kept, and the new model picks up from there. If it does not fit, you are asked to confirm compaction first.

**Compaction** condenses the current conversation into a summary and hands it to the new model. Some earlier details may not be kept. Click **Compact and switch** to continue, or **Cancel** to keep the current model. To keep the details, choose a model with a larger context limit instead. If the session is too long to compact, start a new session.

In a session, the small ring to the left of the model button shows context usage. Hover over it to see the percentage used and the number of tokens. When usage nears the limit, the session compacts automatically.

If a session pauses because the subscription limit ran out or the subscription account disconnected, it continues automatically after you switch models or accounts. See [Model sources](https://docs.aihubmix.com/en/cloud/models/model-sources.md).

> [!TIP]
>
> Open the model picker: `⌃⇧M` (Windows: `Ctrl+Shift+M`). Open the reasoning effort menu: `⌃⇧I` (Windows: `Ctrl+Shift+I`, not available in the web app on Windows). For more shortcuts, see [Keyboard shortcuts](https://docs.aihubmix.com/en/cloud/reference/keyboard-shortcuts.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.