Vertex AutoML Complete Guide (2026)

Vertex AI AutoML is GCP's no-code machine learning service. It can automatically build classification, detection, and prediction models for images, video, text, and tabular data. Teams without deep ML expertise can deliver production-quality models, making it an attractive way to compensate for data scientist shortages.

AutoML Task Catalog

Task	Sub-tasks	Use cases
Tables (Tabular Workflow)	Classification / Regression / Forecasting	Customer churn, demand forecasting
Vision	Image classification / Object detection / Segmentation	Product identification, defect detection
Video	Video classification / Action recognition / Object tracking	Surveillance cameras, sports analytics
Natural Language	Classification / Sentiment analysis / Entity extraction	Review analysis, document classification
Translation	Custom translation models	Industry-specific translation
Recommendations	Recommendation models (Discovery Engine)	E-commerce recommendations

Typical Workflow

Data preparation (CSV / image folders / JSONL)
Create a Dataset (Vertex AI Console)
Labeling (Data Labeling Service is an option)
Start AutoML training (pick a task and set a budget)
Review evaluation metrics (Precision / Recall / Confusion Matrix)
Deploy to an Endpoint (1-click)
Production inference

AutoML Tables Example

from google.cloud import aiplatform as ai

ai.init(project="my-project", location="asia-northeast1")

# Create a Dataset (from a BigQuery table)
dataset = ai.TabularDataset.create(
    display_name="customer-churn",
    bq_source="bq://my-project.dataset.customers",
)

# Training job
job = ai.AutoMLTabularTrainingJob(
    display_name="churn-model",
    optimization_prediction_type="classification",
    optimization_objective="maximize-au-roc",
)
model = job.run(
    dataset=dataset,
    target_column="churned",
    budget_milli_node_hours=1000,  # 1 node-hour
)

# Deploy
endpoint = model.deploy(machine_type="n1-standard-4")
result = endpoint.predict(instances=[{"age": 30, "plan": "premium"}])

Pricing Examples (us-central1, 2026)

Task	Training ($/h)	Prediction ($/h or $/M)
Tables (Classification/Regression)	$21.25	$1.39 per million requests
Tables Forecasting	$21.25	$1.39/M
Image Classification	$3.15	n1-standard-2 hours
Image Object Detection	$18.00	n1-standard-2 hours
Video Classification	$0.46/M frame	$0.10/minute
Text Classification	$1.05	$5/1000 page

AutoML vs. BigQuery ML

Aspect	AutoML	BigQuery ML
Interface	GUI + SDK	SQL only
Supported data	Structured + unstructured	Structured (BigQuery tables)
Model types	Full AutoML lineup	Linear / Tree / k-means / AutoML
Deployment	Vertex Endpoint	In-BQ inference + Vertex Export
Training cost	Node-hours	Data scanned

AutoML Comparison Across Clouds

Aspect	Vertex AutoML	SageMaker Autopilot	Azure ML AutoML
Tabular data	Excellent	Excellent	Excellent
Image	Excellent	—	Good
Video	Excellent	—	—
NL	Excellent	—	Good
Explainability	Feature Importance + Vertex Explainable AI	SHAP	Responsible AI Toolkit

Typical Use Cases

E-commerce: automatic product image classification and tagging
Finance: customer churn prediction (Tables)
Manufacturing: defect detection on inspection images (Vision)
Customer support: inquiry classification (NL)
Video streaming: content classification (Video)
Retail: sales forecasting (Tabular Workflow)

What is AutoML?

A Vertex AI feature that lets you build and deploy machine learning models without code. The flow is fully automated: data upload → automatic feature engineering → model selection → hyperparameter tuning → deployment.

Which tasks does AutoML support?

AutoML Tables (structured data), AutoML Vision (image classification and object detection), AutoML Video (video classification), AutoML Natural Language (text classification, sentiment analysis, entity extraction), and AutoML Translation.

How does it differ from BigQuery ML?

AutoML is GUI-driven with Vertex Endpoint deployment, while BigQuery ML stays entirely in SQL. If your data already lives in BigQuery, prefer BigQuery ML; for unstructured data, go with AutoML.

What about forecasting (time-series prediction)?

AutoML Forecasting handles sales and demand prediction. The Tabular Workflow automatically generates features and uses ensembling.

What is the pricing model?

Node-hour based. For example, image classification is $3.15/h and object detection is $18/h. Vertex AI Tabular Workflow follows the same model.

How is this different from fine-tuning?

AutoML trains a brand-new model from your own data, while fine-tuning adjusts an existing model (Gemini, Llama, etc.). Choose based on the task.

How does it compare to AWS SageMaker Autopilot?

Both are AutoML offerings. Vertex AutoML covers unstructured data like images and video, whereas SageMaker Autopilot is centered on tabular data.

How does it integrate with Vertex AI Workbench?

You can call AutoML programmatically from Workbench (Notebook). In practice, a hybrid GUI + code workflow works best.

Check what you learned with practice questions

Practice with certification-focused question sets

View GCP exam prep page

Author

NicheeLab Editorial Team

NicheeLab editorial team focused on data engineering and cloud certification learning. Content is structured around practical study needs and official exam domains.

Vertex AI AutoML Complete Guide: Tables, Vision, NL, Video, Forecasting