intelligencesupport

AI models

Every AI model is open to everyone. Forbid the ones your organization must not use, choose what Quick, Balanced and Best run, and set defaults.

For
Admins
Updated

The AI Models page decides what your organization’s AI features may run on. Every AI model in the app is open to everyone by default, including the models added later, so there’s no list to build. You forbid the models your organization must not use, choose the model behind each of the three choices every picker opens on (Quick, Balanced and Best), and pick the default model of each module and page. Two smaller cards below hold the Ask Intelligence web search switch and the Playground settings.

How it works

Three layers decide which models a person sees in a model list:

LayerWho sets itWhat it does
Every model in the appThe BearingBridge team adds themOpen to everyone, new models included, with nothing to switch on
Forbidden modelsOrganization admins, on this pageTakes a model away from the whole organization
My modelsEach person, in their SettingsHides a model from that person’s own lists, and nobody else’s

On top of that, each module and each page inside it has a default: the model it runs on when nobody picks one. People pick from three plain choices, Quick, Balanced and Best, whose models you set here, or from the full list of models that aren’t forbidden. The personal switches are covered in My models, and the picker itself in Choosing a model.

Open the page

  1. Open the Admin menu.
  2. Click AI Models.

The page opens on the Forbidden models card, followed by Quick, Balanced and Best, The default model of each module and page, then the Ask Intelligence and Model Playground cards.

Forbid a model

The Forbidden models card lists every model in the app in four folds: Text models, Image models, Video models and Voice input models. Each fold shows how many models are available and, once you forbid some, how many are forbidden. The text models are open when the page loads. Click Show or Hide to open or fold the others. Inside a fold, models are grouped by maker.

The Forbidden models card, with the text models listed by maker and none of them ticked, so every model stays available
The Forbidden models card, with the text models listed by maker and none of them ticked, so every model stays available
  1. Open the fold of the right kind.
  2. Tick the model you want to forbid.

The change saves right away, and the model’s name is struck through. To allow it again, untick it.

A forbidden model is gone for the whole organization:

  • It disappears from every model list: text features, image and video engines, voice input, the Playground and the agents.
  • A run that still asks for it, from a page left open or a choice remembered in a browser, is refused with That model is not available in your organization. Pick another one.
  • An agent set up on it before you forbade it runs on your organization’s Ask Intelligence model instead.
  • Nobody can make it a default. Its line is grayed out in the default menus.

At least one model of each kind always stays allowed. If a tick would forbid the last text, image, video or voice input model, the card refuses it and says Keep at least one text model allowed. (or image, video, or voice input). For images, at least one engine that creates a picture from a prompt has to stay, since the image form opens on one.

The BearingBridge team can also forbid a model for every organization. Such a model shows ticked and grayed out, marked forbidden for every organization, and you can’t untick it.

Tip: You don’t need to forbid a model just to tidy the lists. Each person can hide the models they don’t use in their own Settings, under My models, without changing anything for the others.

Quick, Balanced and Best

Every text model picker opens on three choices, so nobody has to know the models by name: Quick (fast and cheap, for drafts and short posts), Balanced (good quality at a fair cost, for most work) and Best (the strongest, for important work and hard questions). The Quick, Balanced and Best card decides which model each one runs, for your whole organization.

The Quick, Balanced and Best card, each choice with its Runs on menu, here left on the product default
The Quick, Balanced and Best card, each choice with its Runs on menu, here left on the product default
  1. On the choice’s card, open the Runs on menu.
  2. Pick a model. Each one shows its price per million tokens read and written.

The change saves right away and applies to every picker from the next time it loads. The menu’s first option keeps the inherited setting, Global default (…) or Product default (…), with the model and its price in the parentheses: the setting the BearingBridge team applies to every organization, or the app’s own choice. Pick that first option again to go back to inheriting.

A forbidden model is grayed out in these menus and can’t be put on a choice. If you forbid a model a choice already runs, or a person hid it in My models, the choice doesn’t disappear: for that person it runs on the closest model in price they can use, looking among cheaper models first for Quick and stronger ones first for Best.

The default model of each module and page, set on the card below, always sits among the three. In each picker it takes the place of the choice closest to it in price and carries a Recommended here star, so the setting you chose for a page stays the one people are steered to. Two choices never show the same model.

Compare the models on public benchmarks, at the end of the card’s introduction, opens the Model benchmarks page in a new tab: scores, price levels and the cost of a post and of an article for every text model your organization allows. See Choosing a model.

Set a module’s default model

The The default model of each module and page card has one row per module: Ask Intelligence, Content, Ads, Search, Social, CRM (quotes & proposals), Customer Self-Service (public chat) and Email.

Every model in the menus shows two prices: a million tokens read, then a million tokens written. These are the prices billed to you, everything included. A run pays for the tokens it really used, so the same model costs more on a long brief than on a short one. For features that write long texts, the written price weighs most.

The card where each module gets its default model, with the price of a million tokens read and written next to each name
The card where each module gets its default model, with the price of a million tokens read and written next to each name
  1. Find the module’s row.
  2. Open its model menu.
  3. Pick a model.

The change saves right away and applies from the next run. The menu’s first option keeps the inherited setting, and its label says where that comes from:

First optionMeaning
Global default (…)The setting the BearingBridge team applies to every organization
Product default (…)The app’s built-in model
Same as [module] (…)On a page row: the module’s own setting

The model and its price show in the parentheses. Pick that first option again to go back to inheriting.

Defaults are looked up in this order: the page, then the module, then the global page, then the global module, and finally the app default. The global settings belong to the BearingBridge team. You can’t edit them, but you can override them for your organization, and a row you leave alone follows them when they change.

When a default is forbidden

If you forbid the model a row runs on, the row keeps working. An amber line under it says This default is forbidden. It runs on and names the model that stands in: the app’s own default for the row when that one is still allowed, or else the first allowed model. Pick another default to clear the line.

Set a model page by page

A module’s pages can each run on their own default. Use it when quality matters more in one place than another. An Editorial article may deserve a stronger model than a TikTok caption, and short posts often do fine on a cheaper one.

  1. On the module’s row, click Set by page. The button shows how many pages there are and how many you’ve set.
  2. Pick a model on the page’s row.
ModulePages
ContentBrand foundation, Editorial, Commercial, Creative, Corporate & PR, Other (the article workspace), Web (the Web editor)
AdsGoogle Ads, LinkedIn Ads
SearchGoogle Search (SEO), AI Search
SocialCalendar, LinkedIn, Instagram, Facebook, TikTok, X, Re-purposing
Customer Self-ServiceSelf-Service
EmailAnswering Machine, Mass Mailing

Ask Intelligence and CRM have no pages.

Re-purposing, under Social, isn’t a menu page. Its hint reads “Drafts a post again for the other networks”: it’s the model used when a post is taken to other networks with Draft at once or Publish automatically, in the publishing step (see take a post to other networks). People can pick another model there, in an AI model menu. Shorts autopilot, in the Assets Library, runs on it too, to pick a long video’s best moments and frame them.

What people see

Here’s how those settings look to the people running the features.

A compact model button appears next to the button that runs a feature. You’ll find it in Ask Intelligence and across Brand Content text, including the article workspace and the brand foundation form. It’s also on the Social post forms, the Google Ads and SEO keyword tools, the Email pages, the Dashboards builder, and the menu that rewrites a highlighted passage. It shows the current choice, its model and what one typical run costs. Clicking it opens Choose a model: the three cards Quick, Balanced and Best, the default starred Recommended here, then All models, every model your organization allows grouped by maker, minus the ones that person switched off in My models.

  • With a single model left in the list, there’s no button. The form says Runs on [model name].
  • People see what one run of the feature costs on each model, never token prices. The exact cost of each run shows once it’s done, billed on the model that actually ran.
  • A person’s choice is remembered per feature in that browser. On another browser or device, they start from the default again.
  • If you forbid a model someone had picked, their next run is refused with a sentence and the picker reloads without it.
  • If you change the default, everyone who never picked a model, or who picked the starred card, follows the new one.
  • If someone hid the default in My models, their picker opens on their first remaining model instead.

Below the default card, the Ask Intelligence card holds one switch: Allow web search in Ask Intelligence. It’s on by default.

With it on, each person decides question by question, with the web switch on the Ask page. While that switch is on, the Ask model picker offers only models that can search, and the pages read are listed under the answer. With it off, Ask stays on your organization’s own data.

Web search needs a model with its own search. If Ask runs on a model without one, the card says so, and the switch changes nothing until Ask runs on a model that searches. A search is charged like any other paid call and shows in the price of the answer. See Data and web search.

Model Playground

The Model Playground card controls the Playground, where members chat with models directly. Forbidden models aren’t listed on it, since no page can offer them.

Offer every model plugged into the product lets members use every allowed model. Untick it, then tick the ones you want. Tags show which models take files and which have vision. Untick every model to switch the Playground off for the organization.

The Model Playground card, with the models members may chat with and the sampling and allowance settings under them
The Model Playground card, with the models members may chat with and the sampling and allowance settings under them

Under Sampling and allowance:

OptionWhat it does
Lock temperature and max tokensEvery conversation uses your values, and members can’t change them
Temperature (0 to 2)Empty means the model’s own default
Max tokens per answerFrom 256 to 16,000
Monthly allowance per member (USD billed)Counts every Playground answer, file, and image a member billed this month. Once reached, the Playground refuses new messages until next month. Empty means no limit
Allow web searchLets members search the web from the Playground. On by default

Every Playground answer is billed at that model’s rate, and the Playground’s own allowance is on top of the shared-credit rules in Shared credits.

Image, video, and service desk models

Image and video engines follow the same rule as text models: they’re all open unless you forbid them in the Image models or Video models fold. The image and video forms open on the cheapest allowed engine that can do the job, and people can switch engines there. A service desk picks its own model from every text model your organization allows. See Create a desk.

Voice input models, which turn what people say into text in the text boxes, work the same way: all open unless you forbid them in the Voice input models fold. Each person picks one in their own Settings. See voice input.

Who can do this

Changing these settings is free. What people then run on the chosen models is billed as usual; see credits and costs.

TaskWho
Forbid or allow models for the organizationOrganization admins
Set defaults for the organizationOrganization admins
Choose the models behind Quick, Balanced and BestOrganization admins
Allow or block Ask web searchOrganization admins
Set up the PlaygroundOrganization admins
Forbid a model for every organization, change the global defaults and choicesThe BearingBridge team
Hide models from their own listsEach person, in My models
Pick a model when running a featureAnyone running the feature

Organization admins set their own organization only. Whoever opens an organization at sign-up is its admin from the first minute, so they have the page too, even before anyone else joins.

Troubleshooting

“Keep at least one text model allowed.” You tried to forbid the last model of a kind. Allow another one first, then forbid this one.

“That model is forbidden. Allow it again before making it the default.” A forbidden model can’t become a default. Untick it in Forbidden models, or pick another default.

“That model is forbidden. Allow it again before putting it on a choice.” A forbidden model can’t run Quick, Balanced or Best. Untick it in Forbidden models, or pick another model for that choice.

“That model is not available for this module.” The model isn’t in the app’s catalog for that module anymore. Pick another one.

A member still runs on an old model. They picked it themselves in that browser. Ask them to open the model picker and pick the card marked Recommended here, or forbid the model if nobody should use it.

A model you set as a default vanished from your settings. The app stopped offering it. The row falls back to the next level instead of failing.