# Vertex AI Model

> Vertex AI models cost nothing to keep in the Model Registry. Spend starts when a model is deployed to an endpoint and holds serving nodes. ZopNight lists models from the live aiplatform REST API in every GA Vertex region (30+), then links each one to the endpoints and jobs spending money on its behalf.

Source: https://zop.dev/integrations/gcp/vertex-model
Updated: 2026-08-19

---

A Vertex AI model is a trained model artifact registered in the Model Registry. The registry entry is effectively free; cost appears when a model is deployed to an endpoint.

## Registration is free, deployment is not

The Model Registry charges nothing meaningful for holding a model version: the artifact sits in Cloud Storage at ordinary storage rates and the registry row itself carries no meter. Spend begins the moment the model is deployed to an endpoint, into a deployment resource pool, or as the engine behind a batch prediction job, because those are the surfaces that hold machines. Cost materializes at deployment, not registration.

## Fresh model listings from the live API

Cloud Asset Inventory does not index Vertex models, so ZopNight inventories them from the live aiplatform REST API in each GA Vertex region, over 30 locations swept per discovery pass, feeding the same metadata enrichment used for asset-inventory types. The result is a registry view that matches what the API reports right now rather than a lagging snapshot.

## Tracing a model to the money it spends

ZopDev links every model to the endpoints and jobs that spend money on its behalf. That join is the practical value of this row: a model is the connective tissue between the training pipeline that paid to produce it and the serving infrastructure paying to host it, and cost questions about either usually route through the model.

## Registry sprawl and its cheap cleanup

Model-shaped waste is indirect but real: dozens of versions retained when two matter, a superseded model still deployed to an endpoint at a small traffic split, and duplicate registrations of the same artifact across projects. Cleaning the registry costs nothing and shrinks the surface where accidental deployments (the expensive kind of mistake) can happen.

## Model Registry in the console

Google Cloud console → Vertex AI → Model Registry lists models and versions per region. Each version's deploy view shows exactly where it is serving; that is the list to check before assuming a model is dormant.