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

# Selecting an AI Model

> Learn when to leave the model picker on Auto and when to choose a specific model for your task.

Leave the model picker on **Auto** for everyday work. Choose a specific model only when your task needs it, such as very long documents, images, or a workflow already tested on one model.

<video controls preload="none" poster="/images/copilot/video-choose-a-model-poster.webp" src="https://docs.ideaboxai.com/images/copilot/video-choose-a-model.mp4" aria-label="Choosing an AI model in Persona AI: the model picker on Auto, the Basic and Advanced model lists, and the AI Models page in the Admin Console." />

## Start with Auto

**Auto** puts cost efficiency first. It sends each message to the most cost-efficient model that can handle it well:

* **Basic models** handle simple requests: questions and answers, summaries, short drafts, rewriting, tagging, and extraction.
* **Advanced models** are used only when the work needs them: data, tools, connectors, analysis, coding, and multi-step reasoning.

You get the quality you need for the fewest credits.

## Choose a model for your task

If you choose a model yourself, start from your task.

| Your task | Use | Also good | Tier |
| - | - | - | - |
| Mixed or unsure | **Auto** | — | Auto |
| Quick questions, summaries, short drafts | **GPT-6 Luna** | Haiku 5.5 | Basic |
| High-volume extraction, tagging, classification | **GPT-4o-mini** | GPT-6 Luna | Basic |
| Very long documents, contracts, transcripts | **GPT-4.1** | Sonnet 5.5 | Basic |
| Images, screenshots, scanned pages | **GPT-4o** | Sonnet 5.5 | Basic |
| Data analysis, reports, natural-language BI | **Sonnet 5.5** | GLM 5.2 | Advanced |
| Agents that use tools and connectors | **Sonnet 5.5** | Kimi K2.5 | Advanced |
| Coding, scripts, technical configuration | **GPT-6 Sol** | Sonnet 5.5 | Advanced |
| Hardest reasoning, high-stakes decisions | **Opus 5.5** | GPT-6 | Advanced |
| Advanced quality at the lowest credit use | **GLM 5.2** | DeepSeek V3.2 | Advanced |
| A workflow already tested and signed off | **Keep its model** | — | — |

> [!WARNING]
>
> Switching models on a live workflow can change its tone and output format. Test again before you switch a workflow your team relies on.

## Basic models

Basic models are fast and low-cost. Use them for speed and volume.

| Model | Provider | Best for | Credit use |
| - | - | - | - |
| **GPT-6 Luna** | OpenAI | OpenAI's newest fast model. Everyday questions, summaries, and high-volume work. | \$ |
| **Haiku 5.5** | Anthropic | Natural writing at speed. Emails, replies, short drafts, and rewrites. | \$\$ |
| **GPT-4o-mini** | OpenAI | Lowest cost. Simple extraction, tagging, and routing at scale. | \$ |
| **GPT-4.1** | OpenAI | Very long inputs, up to 1M tokens, with precise instruction following. | \$\$ |
| **GPT-4o** | OpenAI | Images and screenshots alongside text. | \$\$\$ |
| **GLM 4.6** | Z.ai | Strong quality for its price. Structured output and light coding. | \$ |
| **GPT-OSS 120B** | AWS Bedrock | Open-weight model at low cost. Routine questions, drafting, and summaries. | \$ |

## Advanced models

Advanced models handle data, tools, and complex work. Quality scores are out of 100.

> [!TIP]
>
> **Best value:** GLM 5.2 scores 93.8, close to the top models, and uses about half the credits of Sonnet 5.5.

| Model | Provider | Best for | Quality | Credit use |
| - | - | - | - | - |
| **Sonnet 5.5** | Anthropic | Default for agents, tool use, and data work. Reliable multi-step execution. | 94.6 | \$\$\$ |
| **Opus 5.5** | Anthropic | Most capable Anthropic model. Hardest reasoning and high-stakes analysis. | 96.2 | \$\$\$\$ |
| **GPT-6** | OpenAI | OpenAI's flagship. Complex research and long, open-ended tasks. | 95.4 | \$\$\$\$ |
| **GPT-6 Sol** | OpenAI | Coding and professional work at mid cost. | 86.6 | \$\$\$ |
| **GPT-5.2** | OpenAI | Previous-generation reasoning. Keep for existing GPT-5 workflows. | 88.0 | \$\$\$ |
| **GPT-5.1** | OpenAI | Previous-generation reasoning at a lower rate than GPT-5.2. | 89.0 | \$\$ |
| **GLM 5.2** | Z.ai | Near-top quality at the best value per credit. Analysis, reports, and coding. | 93.8 | \$\$ |
| **Kimi K2.5** | AWS Bedrock | Agentic tool use at low cost. | 91.3 | \$ |
| **GLM 4.7** | Z.ai | Low-cost Advanced work. Structured tasks and routine analysis. | 90.3 | \$ |
| **DeepSeek V3.2** | AWS Bedrock | Lowest-cost Advanced model. | 89.2 | \$ |
| **Kimi K2 Thinking** | AWS Bedrock | Step-by-step reasoning on a budget. | 88.2 | \$ |

**Credit use** is per 1M tokens:

| Symbol | Credits per 1M tokens |
| - | - |
| \$ | Under 100 |
| \$\$ | 100–250 |
| \$\$\$ | 250–400 |
| \$\$\$\$ | Over 400 |

## Manage models in the Admin Console

Admins choose which models people in the workspace can use. In the **Admin Console**, under **Workspace**, select **AI Models**.

![The AI Models page in the Admin Console, listing Basic models such as Haiku 5.5, GPT-OSS 120B, GPT-4o-mini, and GPT-6 Luna, each with its quality score, price, availability, a Manage access link, and an on/off toggle.](https://mintcdn.com/ideabodxai/JcuZvIXfno22mZa-/images/copilot/ai-model-management.webp?fit=max\&auto=format\&n=JcuZvIXfno22mZa-\&q=85\&s=6d7310308fb41e32748c9d0a69f3d47d)

Models are grouped into **Basic** and **Advanced**. Each model shows its provider, quality score, price per 1M tokens, the date it was rated, and who it's available to.

* **Turn a model on or off:** Use the toggle. Turned-off models are hidden from everyone's model picker, and Auto skips them.
* **Choose who can use a model:** Select **Manage access**. See [Choose who can use a model](#choose-who-can-use-a-model).
* **Update quality scores:** Select **Refresh scores**.

> [!WARNING]
>
> Keep at least one model turned on in each category, Basic and Advanced.

### Choose who can use a model

**1. Open Manage access**

On the model's card, select **Manage access**. The **Manage access** dialog opens with **Role access** at the top.

**2. Set role access**

Under **Role access**, select **Access** for each role that can use the model. Everyone in a selected role gets access by default. Select **Own key** to let a role use its own provider key for the model. Use **Filter roles** to find a role.

**3. Set user access (optional)**

Scroll down to **User access**. Select or clear **Access** for individual users. These settings override role access. Users who get access from a role show **Inherited**.

**4. Save**

Select **Save changes**.

Auto only picks models a person is allowed to use. Everyone must keep at least one model in each category, so you can't save a change that would leave a role or user with none.

The list shows the models available to your workspace.

## Managed credits or your own key

| | Managed (default) | Bring your own key |
| - | - | - |
| **Models** | Every model on this page, ready to use. | Use your own Anthropic, OpenAI, Z.ai, or AWS key. |
| **Setup** | None. | A user with the System Admin role sets it up. |
| **Billing** | Usage comes from your IdeaBoxAI credit pool, on one IdeaBoxAI invoice. | Your provider bills you directly. No credit cap. |
| **Limits** | Your plan's concurrent-user limits apply. | Your plan's concurrent-user limits still apply. |

> [!NOTE]
>
> This page reflects models and rates as of November 2026. Models, quality scores, and provider prices change. IdeaBoxAI applies the rate in effect when you use a model.


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