# Vertex AI Tuning Job

> Vertex AI tuning jobs bill either for tuning compute or per training token, depending on the model being fine-tuned. ZopNight inventories tuning jobs through its live aiplatform sweep of 32 Vertex regions so repeated fine-tuning iterations on large foundation models appear alongside the rest of your ML spend.

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

---

A tuning job fine-tunes a foundation model on your data, billed for tuning compute or per training token depending on the model. Repeated tuning iterations on large models add up quickly.

## Tuning compute vs per-token billing

Two meters exist depending on what is being tuned. Managed foundation-model tuning bills per training token, so dataset size and epoch count are the bill. Other tuning paths bill for the compute the job holds, like any training workload: machine specification, accelerators, duration. Either way the job terminates when done and leaves no standing cost, which makes the spend profile a series of spikes rather than a flat line, easy to overlook in a monthly view and painful in aggregate.

## Fine-tuning spend in ZopNight's inventory

ZopNight inventories tuning jobs through its live aiplatform sweep across 32 Vertex regions. Cloud Asset Inventory's rows for transient jobs go stale, and the live listing keeps completed jobs from masquerading as active ones. Fine-tuning spend then lands alongside the rest of ML cost attribution, so the cumulative price of iteration is visible instead of buried. Tuning jobs are not schedulable; they run to completion on their own.

## Iteration cost on foundation models

The characteristic leak is not one expensive job but many: hyperparameter exploration re-tuning the same model dozens of times; teams tuning on the full corpus when a fraction would have validated the approach; and abandoned experiments whose tuned models nobody ever deploys, turning the entire token spend into retroactive waste. Iterations per shipped model is the honest metric to watch.

## Tuning jobs in the Vertex AI console

Google Cloud console → Vertex AI Studio → Tuning lists tuning jobs with state and the base model. Cross-reference that list against deployed models to see which tuning spend ever reached production. A long tuning history feeding a short deployment list is the clearest sign that iteration discipline, not machine choice, is where the money is going.