Models and spending
Choose which AI models your organization may use, set defaults for the organization, a service or a workspace, and cap what Fibo spends.
On this page6 sections
The Models page#
Open Settings, then Models. The page lists every model available to your organization, grouped by provider, with its prices: input and output per million tokens, and cached input where it applies.
Everyone can read this page and star models as favorites. Organization admins change the settings on it.
The list is kept current automatically. It leaves out models that are deprecated, and models whose provider offers no option to keep your data from being retained or used for training.
Allow or turn off models#
Each provider has a switch that allows all of its models, and each model has its own switch. Turn a model off and nobody in the organization can pick it.
Rules can be narrowed at three levels, and each level can only turn models off, never turn back on what a level above turned off.
| Level | Where | Who changes it |
|---|---|---|
| Organization | Settings, Models | Organization admins. |
| Service | The service's Configuration page | Organization admins, and people who manage the service's access. |
| Workspace | Models in the workspace's left bar | The workspace's owner. |
A model turned off at a higher level shows as off, with a lock, at the levels below.
A change reaches a workspace the next time it is opened.
Default model and effort#
The default is the model a conversation uses when the person has not chosen one. It is shown in the model picker with a Default tag that says where it comes from.
Defaults can be set at the same three levels. The most specific one wins: the workspace's, then the service's, then the organization's. With none set, the platform's default applies, which is currently Claude Haiku 5.5.
Default effort works the same way. It sets how much thinking the model does before answering: off, low, medium, high or max.
A model that is turned off cannot be chosen as a default.
What people choose for themselves#
In a conversation, each person can pick any allowed model and an effort level. Their choice is kept for that workspace in their browser. Favorites follow them across devices, up to 10.
See Conversations.
Fibo#
Two settings on Settings, Models apply to everyone's Fibo. Only organization admins see them.
Fibo's model is the model every person's Fibo answers with. People do not choose a model in the Fibo panel.
Fibo spending limit caps what everyone's Fibo may spend together in a calendar month. Left empty, there is no limit.
When the limit is reached:
- people can read their Fibo conversation but cannot send until the limit is raised or the next month starts;
- Fibo cannot compose its notification messages, so people get a short fixed message instead that points them to Home;
- admins see a notice on Home and on Settings, Fibo, with a button to raise the limit.
The limit can be passed slightly by replies that were already under way when it was reached.
The limit covers Fibo only. When Fibo asks a workspace's agent a question, that answer is paid for by the workspace. The work agents do in workspaces is not capped by this setting.
See what work costs#
- In a conversation, each reply shows its cost, and the header shows the conversation's total.
- A workspace's Details shows its cost so far.
- A skill's Track record shows cost per use and per version.