Model details

Every catalog model has its own page. It answers one question for one tenant: what this model is, and how it is doing for that tenant. Open it by selecting a model’s name on Build › Models.

Who can open it

Role	What they can do
Admin	Read the page, deploy and undeploy the model, read the model server’s log
Builder	Read the page, including the deployment’s steps and conditions
Auditor	Read the page, including the deployment’s steps and conditions
Reviewer	No access
End user	No access
Platform admin	Holds every capability: also sees the list price and the model registry, edits the LLMaaS offering, and removes the model from the catalog

A section whose data you cannot read is left out, not shown empty.

What you see

The breadcrumb back to Models, the model’s display name and its description. On the right, Deploy when you may deploy it and nothing runs, and, for a platform admin, a menu with Remove from catalog.

Below the header, the tenant picker names the tenant the page describes. It is hidden when you belong to one tenant.

At a glance

Four tiles:

Tile	What it shows
Deployment	The deployment status for the tenant, such as Active or Not deployed. For an external model the tile reads Serving, with the provider and its data residency
Tokens this month	Tokens the tenant sent through this model this month. Links to Usage
Requests this month	Requests the tenant sent to it this month
Agents that use it	How many of the tenant’s agents answer with it

A tile that could not be read says so rather than showing zero.

Deployment for your tenant

For a model the platform runs, the section Deployment for tenant shows, when the model is deployed:

Status, with the reason when it failed or is not served.
Where it runs: Project (environment) on cluster, or The platform’s own namespace. The Namespace follows.
Replicas, when more than one runs.
Serving backend, Deployed and Last change.
The rollout steps, from Plan checked to Ready to serve, each marked complete, current, failed or waiting.
Model server log, for tenant admins: the newest 200 lines, refreshed every 5 seconds while the deployment rolls out. Choose the Replica and Container, Pause auto-refresh, Refresh, or Copy the log. When the server keeps restarting, the log shows the last lines of its previous run, which say why it stopped.
Conditions: a table of Condition, State (Met, Not met, Not known yet), Reason and Since.

The conditions, in order:

Condition	Met when
Plan is valid	The plan passed its checks
Weights are available	The model’s weights were found
Placed in a namespace	The workload was placed in the project’s namespace
Rolled out	Every replica of the model server is running
Granted to the tenant	The tenant may call the model
Routable at the gateway	The gateway publishes the model
Ready	All of the above

A failed deployment’s cause is its first unmet condition.

When nothing runs, the section reads Not deployed for tenant. Tenant admins deploy from the header. See Deploy a model.

Served by a cloud provider

For an external model, Served by a cloud provider replaces the deployment section. It says the model is not deployed and that prompts and responses are sent to the provider and processed there, then lists Provider, Model at the provider, Data residency, Tool calling (for an LLM) and Context window.

Agents that use it

Each of the tenant’s agents that answers with this model, linked to the agent, with its description. When there is none: No agent of tenant answers with this model.

About this model

The model’s card as published with it, or its description. When there is neither: No description available.

Facts

What is known of the model, with empty rows left out: Author, Type, Task, Parameters, Download size, License, Gated, Downloads (all-time), Likes, Released, and View on Hugging Face for a model added from the hub.

LLMaaS

Whether the platform offers the model as a managed endpoint (Offered or Not offered) and its Token weight: how many tokens each token of this model counts as against a budget. A platform admin also sees the GPU footprint, the selected tenant’s Subscription, Daily budget, Monthly budget and Usage this month (weighted), Edit offering, and a link to change the tenant’s subscription and budgets.

List price and model registry (platform admins)

List price shows Input, per million tokens and Output, per million tokens from each live price book, with a link to Price books in Billing.

Model registry shows the packaged weights held for the model: Status, Revision, Size and Imported, with Open in the model registry. When none are held, a cluster fetches the weights from their source.

Empty and error states

This model is not in the catalog. It may have been removed. Go to Models returns to the catalog.
This model’s details could not be loaded. Retry from the message.
What is deployed could not be loaded, The agents could not be loaded, The price list could not be loaded, The registry could not be read: each section reports its own failure and offers a retry where one helps.

What you can do

Deploy the model

Select Deploy in the header.
Follow the wizard. See Deploy a model.

Find out why a model is not serving

Open the model’s page for the tenant.
Read Status and the rollout steps under Deployment for tenant. The failed step carries the reason.
Open Conditions for every reason and the time it changed.
As a tenant admin, read the Model server log.

Undeploy the model

Select Undeploy on Deployment for tenant.
Confirm in Undeploy model.

Requests to the model fail from then on, and the tenant’s grant on it is revoked.

Remove the model from the catalog (platform admins)

Open the menu in the header and select Remove from catalog.
Confirm.

No tenant can deploy it afterwards. It is refused while any tenant still runs the model.

Limits

Rule	Value
Log lines per read	The newest 200
Log refresh while rolling out	Every 5 seconds
Status refresh while the deployment changes	Every 3 seconds