Choose the AI model

The AI model is the engine behind every answer Centrus gives. It decides how many message credits an answer uses, how much thinking Centrus can do before it replies, and which kinds of file it can make for you. Every model in the list can search your files and the web, create images, and read the PDFs, spreadsheets and photos you attach.

app.centrus.ai/app/chat

Chat

Start conversations with Centrus

Start new chat

Search chats, messages, and people...

Folders

Sort by:DateNewestBulk Actions

Your personal chat history, excluding agent-generated threads.

John MartinContracts09 Oct 09:14

Riverside House P1 response timesUnder the Riverside House SLA, a priority 1 fault...

John MartinTechnical08 Oct 16:20

Ideal Logic boiler fault L2L2 means the boiler failed to light. Check the gas...

John MartinPricing08 Oct 11:02

Out of hours call-out ratesOut of hours attendance is charged at £145.00...

John MartinSOPs07 Oct 07:50

RAMS for Mill Lane roof accessHere is the RAMS template, filled in for...

John Martin

Riverside House P1 response times

last message sent: 2 minutes ago

Contracts

AI Model

1

Share chat

Download chat

Swap tags

Archive

Rename

What response time do we owe Riverside House for a priority 1 fault, and is there a penalty if we miss it?

09:12

Under the Riverside House SLA, a priority 1 fault needs an engineer on site within 4 hours, at any time of day. If we miss it, the client can claim a service credit of 5% of that month's maintenance charge for each failure.

Sources:

• • Riverside House SLA.pdf

09:14

2Type a message...

Centrus may make mistakes. Check important info.

1. 1AI Model, in the gear menu of a chat you own, changes your own model.
2. 2The reasoning button on the message bar sets how much the model thinks before it answers.
3. 3Settings holds your own model on the Account tab. Admins set the organisation's model on the AI Model tab.

There are two settings, and Centrus checks them in order each time it answers:

• Personal AI Model is yours alone. Anyone can set it, whatever their role, and it applies to every chat you own.
• Organisation AI Model is set by an admin and applies to everyone who has not chosen a model of their own.
• If neither is set, Centrus uses GPT-5.6 Terra, the model the list marks as recommended.

The quickest way to switch is from any chat you own. AI Model only appears in the gear menu of your own chats: in a chat someone has shared with you, the menu offers Download chat and nothing else.

1. 1Open one of your chats and click the gear at the top right of the chat.
2. 2Click AI Model. The Change AI Model window opens, showing your current choice.
3. 3Open the list and click a model, or click Use organisation AI model to follow your organisation's choice again.
4. 4Wait a moment while the page reloads. The chat opens again, with the message bar's buttons matched to the new model.

The window says the model is for this chat, but the choice is saved to your account. It changes the model for every chat you own, including your WhatsApp and Microsoft Teams conversations, and it is the same setting as Personal AI Model in Settings.

Once you choose, the list locks while Updating model capabilities... shows, two confirmations appear, AI model updated. Page will refresh to update features. and Updated, and the page reloads under a second later. If the change cannot be saved, Failed to update AI model appears and your model stays as it was. Choosing the model you already have just closes the window.

You can also set your own model without opening a chat. Everyone can do this, whatever their role.

1. 1Click Settings in the sidebar. The Account tab opens.
2. 2Scroll down to Personal AI Settings.
3. 3Choose a model under Personal AI Model, or choose Use organisation AI model to follow your organisation's choice.
4. 4Scroll to the foot of the page and click Save changes. A notification reading Updated confirms it.

The information icon beside Personal AI Model explains: "If set, this will override any default organisation AI model for your account only, allowing you to customise your own AI experience." Nothing changes until you click Save changes, and Discard changes puts back what was saved before. Settings does not reload the page, and your next message uses the new model.

Admins have the same choice on the AI Model tab too, under Personal AI Model, as described in Set your organisation's model.

The list shows each model's name and four marks, so you can compare what they do at a glance. It does not show credits: the information icon beside each name does.

Every model in the list has all four ticks except GPT-4.1, which has a red cross for Reasoning.

The details window gives OpenAI's description of the model, what it can do, the date it was released and its Message credits per message. The description mentions a context window, which is how much text the model can take in at once, and a knowledge cutoff, which is how recent its general knowledge is. The cutoff does not stop Centrus using your files, which it searches every time you ask, or finding recent information with web search.

These are the 15 models in the list, in the order it shows them. The credits are for one answer on the web, before any of the extras described in the next section.

• Reasoning gives the range of levels on the reasoning button for that model. GPT-4.1 has no reasoning button. The levels are explained in Reasoning effort and temperature.
• Files it can make: every model from GPT-5.2 onwards can make every kind of file, including PDFs, Word documents and PowerPoint presentations. GPT-5.1, GPT-5, GPT-5 Mini, GPT-5 Nano and GPT-4.1 make spreadsheets, CSV files and PNG charts only. See Files Centrus makes for you.
• Files you attach are read the same way by every model in the list. See Which AI models read which files.

A few older models have been taken out of the list: GPT-4o, GPT-4o Mini, GPT-4.1 Mini, GPT-4.1 Nano, O1, O3, O3 Mini and O4 Mini. Anyone, or any organisation, already set to one of them keeps it, and each answer uses 1 credit. Once you change to another model, you cannot choose it again. O1, O3, O3 Mini and O4 Mini cannot read photos or make files.

Every answer uses message credits from your organisation's allowance. The model sets the starting cost, and three things can add to it:

• The model: 1 credit for most models, 2 for GPT-5.6 Terra and GPT-5.2 Pro, 4 for GPT-5.6 Sol and 10 for GPT-6 Astra.
• The document sandbox: 2 more when Centrus builds a file for you, once per answer however many files it makes. See Files Centrus makes for you.
• Images: each image made with the picture button adds 1 to 10 credits, depending on its quality. Auto and medium quality add 2. See Web search and images.
• WhatsApp: 1 more for every answer sent on WhatsApp. Answers in Microsoft Teams cost the same as on the web.

The reasoning level does not change the cost: an answer at Max uses the same credits as one at None. An answer that fails before any reply reaches you is not charged.

The credits come from one allowance shared by the whole organisation, so a person on GPT-6 Astra uses it five times as fast as a colleague on GPT-5.6 Terra. See Credits for how the allowance works.

For most property maintenance, facilities and field service teams, GPT-5.6 Terra is the right place to start. It is the model Centrus recommends, it can make every kind of file, and it offers every reasoning level from None to Max. Change it when you have a reason to.

• GPT-5.6 Luna suits people who send lots of short questions, such as engineers asking on WhatsApp for a fault code, a part number or a site's access times. It can do everything Terra can for 1 credit an answer, and OpenAI describes it as its cost-efficient GPT-5.6 model for high-volume work.
• GPT-5.6 Sol and GPT-6 Astra suit the occasional heavy job, such as checking a long maintenance contract against its SLA or building a full report from many files. OpenAI describes Sol as its flagship model for complex professional work, and Astra as its most capable, for the hardest end-to-end work. At 4 and 10 credits an answer, they are best chosen for a task and changed back afterwards.
• Older models need a check first. GPT-5.5, GPT-5.4 and GPT-5.2 also cost 1 credit and make every file type. GPT-5.1, GPT-5 and GPT-4.1 cost 1 credit too, but make only spreadsheets, CSV files and charts, so they cannot hand back a PDF report or a Word document.

Before moving to a model that costs more, try a higher reasoning level on the one you have, as it uses no extra credits. The best test is your own work: ask the same few everyday questions on Terra and on the model you are considering, then compare the answers and the credits used.

Most models can think a question through before they answer. The reasoning button on the message bar, shaped like a brain, sets how much: click it and choose a level under Reasoning effort. Higher levels take longer and suit harder questions, while lower levels reply sooner.

• GPT-5.6 Sol, Terra and Luna: None (Fastest), Low, Medium, High, Extra High and Max (Most thorough). They start at None.
• GPT-6 Astra: Low (Recommended) up to Max (Most thorough). It always reasons, so None is not offered, and it starts at Low.
• GPT-5.5 and the GPT-5.4 models: None (Fastest) up to Extra High, starting at None.
• GPT-5.2: None (Fastest, supports temperature) up to Extra High. GPT-5.1 is the same up to High. Both start at None.
• GPT-5.2 Pro: Low, Medium (Recommended), High and Extra High, starting at Medium.
• GPT-5, GPT-5 Mini and GPT-5 Nano: Low, Medium (Recommended) and High, starting at Low.
• GPT-4.1 does not reason, so the button is not shown.

The level you pick is not saved. It applies to the messages you send until the page reloads, then returns to the model's starting level, so it also resets every time you change model. On a phone, the button is in the tools menu at the left of the message bar and opens Select Reasoning Level. WhatsApp and Microsoft Teams have no reasoning button, so every answer there uses the model's starting level. When Centrus builds a file in its document sandbox, it uses at least Medium for that work, whatever level you chose.

Temperature controls how much the wording of answers varies: lower is more consistent, higher more varied. Only GPT-5.2, GPT-5.1 and GPT-4.1 accept it, and GPT-5.2 and GPT-5.1 use it only when reasoning is set to None. With one of those as your personal model, a Personal Temperature slider from 0 to 1 appears beneath it in Settings, starting at 0.50. For every other model the page reads Temperature is not supported for the selected reasoning model.

Admins choose the model everyone uses unless they pick their own. The AI Model tab in Settings is only shown to admins, and only an admin can save it. Roles are covered in Users and roles.

1. 1Click Settings in the sidebar, then the AI Model tab.
2. 2Click Show advanced, below the two context boxes.
3. 3Choose a model under Organisation AI Model.
4. 4Scroll to the foot of the page and click Save changes. A notification reading Updated confirms it.

The information icon beside it explains: "Organisation AI Model sets the default AI model for all users in your organisation. This model will be used for everyone unless a user selects a personal AI model, in which case their personal choice will override this default for their account only." The context boxes at the top of the page are covered in Personal AI settings.

Reset to default does more than its name suggests. It replaces your Organisation Context with Centrus's standard wording, empties your Personal Context, sets the organisation's model to GPT-4o, an older model that is no longer in the list, and puts your own settings back to the organisation's. Nothing is saved until you click Save changes, so click Discard changes if you pressed it by mistake.

An organisation created through the sign-up page starts on GPT-4.1, not on the recommended GPT-5.6 Terra. GPT-4.1 cannot reason and makes only spreadsheets, CSV files and charts, so check this setting when you first set Centrus up.

Each answer comes from the model of the person who owns the chat, worked out again every time Centrus replies. So a change applies from the next message, and earlier answers stay as they were.

Organisation AI Model

Set by John in Settings, AI Model

GPT-5.6 TerraRECOMMENDED▼

John Martin

Admin

Personal AI Model

GPT-5.6 Sol▼

Answers in their chats

GPT-5.6 Sol4 credits each

His own choice comes first

SP

Sarah Price

Contracts manager

Personal AI Model

Use organisation AI model▼

Answers in their chats

GPT-5.6 Terra2 credits each

She follows the organisation

DH

Dylan Hughes

Engineer

Personal AI Model

GPT-5.6 Luna▼

Answers in their chats

GPT-5.6 Luna1 credit each

His own choice comes first

• Your Personal AI Model changes only your chats. Nobody else's answers change.
• The Organisation AI Model changes the answers of everyone who has left their own choice on Use organisation AI model. People who have picked a model keep theirs.
• Existing chats carry on from where they were. You can switch model in the middle of a conversation and keep asking about it.
• WhatsApp and Microsoft Teams conversations are your chats too, so they use your model.
• Shared chats keep their owner's model. People a chat is shared with can read it but cannot send to it. See Share a chat and see team history.
• Agents can be given a model of their own in the Agent Builder, which takes the place of these settings for that agent. See Open Agent Builder.

The Use organisation AI model option does not name the model your organisation uses. If you need to know, ask an admin.

On a phone, the gear menu and the Change AI Model window work the same way. The list opens nearly as wide as the screen, the four marks follow each name, and the information icon moves to the end of the row. The RECOMMENDED badge can cut GPT-5.6 Terra's name short; tap the information icon to see it in full.

09:41

Centrus

Change AI Model

Select an AI model to use for this chat. Different models have different capabilities such as web search, file attachments, and reasoning.

Note: The page will refresh after changing models to update all available features.

GPT-5.6 TerraRECOMMENDED▼

Use organisation AI model

GPT-6 Astra

GPT-5.6 Sol

GPT-5.6 TerraRECOMMENDED

GPT-5.6 Luna

Tap for details

GPT-5.5

GPT-5.4

GPT-5.4 Mini

GPT-5.4 Nano

Tap a model to choose it, and the page reloads as it does on a computer. Personal AI Model on the Account tab works on a phone too, with Save changes at the foot of the page.

Before each answer, Centrus checks that your organisation has enough credits left to cover the model, plus 1 on WhatsApp and the cost of an image if you asked for one. If it does not, the message is turned away.

• A model that costs more can be turned away while a cheaper one still works. With 3 credits left, a question to GPT-5.6 Sol, at 4 credits, is refused, while GPT-5.6 Luna, at 1, still answers.
• The check leaves out the document sandbox, so an answer that builds a file can take your organisation's usage 2 credits past its allowance.
• There is one model per person. You cannot set a different model for a single chat. To use another model for one task, change it, then change it back.
• Changing model from a chat reloads the page. Changing it in Settings does not.
• The reasoning level resets to the model's starting level whenever the page reloads.
• Older models taken out of the list keep working for anyone already set to them, but cannot be chosen again once changed.

When the allowance runs out, an admin can buy more message credits. See Credits and Plan and usage.