Register Vertex AI Models Directly in BigQuery SQL

Registering Vertex AI Remote Models in BigQuery

A data team is expanding their data warehouse capabilities by integrating external machine learning capabilities into BigQuery. They have set up a BigQuery Cloud resource connection named my_connection in the us multi-region and granted the associated service account the required Vertex AI permissions.

Which Data Definition Language (DDL) statement should the team run in BigQuery to register a remote model that connects through this external connection?

CREATE OR REPLACE EXTERNAL TABLE my_project.my_dataset.remote_llm WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm IMPORT FROM CONNECTION my_project.us.my_connection OPTIONS (MODEL_TYPE = 'VERTEX_AI');

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm REMOTE WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM my_project.my_dataset.training_data;

Explanations

CREATE OR REPLACE EXTERNAL TABLE `my_project.my_dataset.remote_llm` WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

What This Option Proposes

This statement attempts to create an External Table (CREATE OR REPLACE EXTERNAL TABLE) pointing to a machine learning endpoint.

Why It Doesn't Meet the Requirement

Wrong object type (✗): External tables in BigQuery are designed to query structured and semi-structured datasets stored in external storage systems like Cloud Storage, Bigtable, or Google Drive, not machine learning inference services.
Unsupported endpoint option (✗): External table definitions take options such as format, uris, and sheet_range. They do not support machine learning parameters such as ENDPOINT or Vertex AI foundation models.
Incompatible with ML functions (✗): Functions such as ML.GENERATE_TEXT or ML.PREDICT require a model resource (MODEL), not a table resource.

Key Characteristics of External Tables

Used to query external tabular files (CSV, Parquet, ORC, AVRO, JSON) in place without loading them into BigQuery native storage.
Created with clauses like WITH CONNECTION to access storage buckets with delegated permissions.

This Answer Would Be Correct If

The requirement were to query Parquet or CSV data files stored in an external Cloud Storage bucket via a BigQuery Cloud resource connection.

View source documentation

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` IMPORT FROM CONNECTION `my_project.us.my_connection` OPTIONS (MODEL_TYPE = 'VERTEX_AI');

What This Option Proposes

This option attempts to register an external model using an invalid IMPORT FROM CONNECTION SQL clause combined with a fictional MODEL_TYPE = 'VERTEX_AI' option.

Why It Doesn't Meet the Requirement

Invalid DDL syntax (✗): BigQuery ML does not support an IMPORT FROM or IMPORT FROM CONNECTION statement. Remote models must be declared using the REMOTE WITH CONNECTION clause.
Invalid model type setting (✗): MODEL_TYPE is not used with an argument of 'VERTEX_AI' to establish a remote connection; instead, remote models configure the connection via REMOTE WITH CONNECTION and set the target via ENDPOINT or REMOTE_SERVICE_TYPE.

Key Characteristics of Model Import vs. Remote Models

Imported models in BigQuery ML refer to models trained outside BigQuery (such as ONNX or TensorFlow models) and imported from Cloud Storage into BigQuery storage for local execution.
Remote models remain hosted on external services like Vertex AI and are invoked via API calls rather than imported into BigQuery storage.

This Answer Would Be Correct If

BigQuery ML utilized an IMPORT FROM CONNECTION keyword syntax for loading model binaries directly into BigQuery tables, which is not part of GoogleSQL standard syntax.

View source documentation

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` REMOTE WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

What BigQuery Remote Model DDL Is

A remote model in BigQuery ML acts as a reference to a machine learning model or service hosted outside BigQuery's native execution engine, such as a foundation model deployed on Vertex AI or a Cloud AI service. Defining a remote model allows SQL developers and data analysts to run inference directly against powerful external models using standard BigQuery SQL statements without extracting data or writing bespoke Python client code.

An annotated breakdown of the CREATE MODEL remote DDL showing the model name, REMOTE WITH CONNECTION clause, and ENDPOINT option and what each part does.

How It Meets the Requirements

Remote model declaration (✓): Uses the standard syntax CREATE OR REPLACE MODEL followed by the dataset and model resource name.
Connection clause (✓): Explicitly specifies REMOTE WITH CONNECTION paired with the fully qualified connection path (my_project.us.my_connection), allowing BigQuery to use the connection's delegated service account credentials.
Target model specification (✓): The OPTIONS clause specifies the ENDPOINT (or model name) pointing to the corresponding Vertex AI service endpoint in the designated location.

Key Advantages and Capabilities

Zero data processing costs during creation: Creating a remote model metadata definition processes 0 bytes in BigQuery and incurs no query compute costs.
Unified governance: Access to Vertex AI is mediated through a single delegated BigQuery Cloud resource connection service account rather than managing individual user keys.
In-place SQL inference: Once defined, users can invoke the remote model directly inside SQL queries using functions like ML.GENERATE_TEXT or ML.PREDICT.
Cross-region compatibility: Supports regional and multi-regional datasets pointing to Vertex AI regional endpoints.

Why This Is Optimal

This statement uses the exact GoogleSQL DDL syntax designed for registering external and remote models. It encapsulates the connection identifier and target endpoint parameters cleanly within the standard REMOTE WITH CONNECTION definition.

View source documentation

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM `my_project.my_dataset.training_data`;

What This Option Proposes

This statement constructs an internal model training query using an AS SELECT query block with a TRANSFORM clause, attempting to pass the connection as a feature transformation parameter.

Why It Doesn't Meet the Requirement

Invokes model training instead of remote registration (✗): The AS SELECT ... clause is used for training internally trained models (such as Linear Regression, Logistic Regression, or K-means clustering). Remote models do not train on rows in BigQuery and do not take an AS SELECT training query.
Misuses the TRANSFORM clause (✗): The TRANSFORM clause in BigQuery ML is used to specify manual feature preprocessing functions (e.g., ML.STANDARD_SCALER, ML.BUCKETIZE) during training, not for specifying external connection parameters.
Missing REMOTE WITH CONNECTION (✗): Omits the mandatory keyword REMOTE WITH CONNECTION required to link the model resource to BigQuery's connection framework.

Key Characteristics of Internally Trained Models

Require an AS SELECT query to ingest training features and target labels from BigQuery tables.
Process bytes and consume BigQuery computing slots during model training.
Fully managed and executed within the BigQuery infrastructure rather than external API endpoints.

This Answer Would Be Correct If

You were training a built-in supervised BigQuery ML model (such as a logistic regression or boosted tree model) and applying custom feature preprocessing rules using SQL functions.

View source documentation

Last updated October 23, 2026

Registering Vertex AI Remote Models in BigQuery

Which Data Definition Language (DDL) statement should the team run in BigQuery to register a remote model that connects through this external connection?

CREATE OR REPLACE EXTERNAL TABLE my_project.my_dataset.remote_llm WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm IMPORT FROM CONNECTION my_project.us.my_connection OPTIONS (MODEL_TYPE = 'VERTEX_AI');

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM my_project.my_dataset.training_data;

Explanations

CREATE OR REPLACE EXTERNAL TABLE `my_project.my_dataset.remote_llm` WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

What This Option Proposes

This statement attempts to create an External Table (CREATE OR REPLACE EXTERNAL TABLE) pointing to a machine learning endpoint.

Why It Doesn't Meet the Requirement

Wrong object type (✗): External tables in BigQuery are designed to query structured and semi-structured datasets stored in external storage systems like Cloud Storage, Bigtable, or Google Drive, not machine learning inference services.
Unsupported endpoint option (✗): External table definitions take options such as format, uris, and sheet_range. They do not support machine learning parameters such as ENDPOINT or Vertex AI foundation models.
Incompatible with ML functions (✗): Functions such as ML.GENERATE_TEXT or ML.PREDICT require a model resource (MODEL), not a table resource.

Key Characteristics of External Tables

Used to query external tabular files (CSV, Parquet, ORC, AVRO, JSON) in place without loading them into BigQuery native storage.
Created with clauses like WITH CONNECTION to access storage buckets with delegated permissions.

This Answer Would Be Correct If

The requirement were to query Parquet or CSV data files stored in an external Cloud Storage bucket via a BigQuery Cloud resource connection.

View source documentation

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` IMPORT FROM CONNECTION `my_project.us.my_connection` OPTIONS (MODEL_TYPE = 'VERTEX_AI');

What This Option Proposes

This option attempts to register an external model using an invalid IMPORT FROM CONNECTION SQL clause combined with a fictional MODEL_TYPE = 'VERTEX_AI' option.

Why It Doesn't Meet the Requirement

Invalid DDL syntax (✗): BigQuery ML does not support an IMPORT FROM or IMPORT FROM CONNECTION statement. Remote models must be declared using the REMOTE WITH CONNECTION clause.
Invalid model type setting (✗): MODEL_TYPE is not used with an argument of 'VERTEX_AI' to establish a remote connection; instead, remote models configure the connection via REMOTE WITH CONNECTION and set the target via ENDPOINT or REMOTE_SERVICE_TYPE.

Key Characteristics of Model Import vs. Remote Models

Imported models in BigQuery ML refer to models trained outside BigQuery (such as ONNX or TensorFlow models) and imported from Cloud Storage into BigQuery storage for local execution.
Remote models remain hosted on external services like Vertex AI and are invoked via API calls rather than imported into BigQuery storage.

This Answer Would Be Correct If

BigQuery ML utilized an IMPORT FROM CONNECTION keyword syntax for loading model binaries directly into BigQuery tables, which is not part of GoogleSQL standard syntax.

View source documentation

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` REMOTE WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

What BigQuery Remote Model DDL Is

How It Meets the Requirements

Remote model declaration (✓): Uses the standard syntax CREATE OR REPLACE MODEL followed by the dataset and model resource name.
Connection clause (✓): Explicitly specifies REMOTE WITH CONNECTION paired with the fully qualified connection path (my_project.us.my_connection), allowing BigQuery to use the connection's delegated service account credentials.
Target model specification (✓): The OPTIONS clause specifies the ENDPOINT (or model name) pointing to the corresponding Vertex AI service endpoint in the designated location.

Key Advantages and Capabilities

Zero data processing costs during creation: Creating a remote model metadata definition processes 0 bytes in BigQuery and incurs no query compute costs.
Unified governance: Access to Vertex AI is mediated through a single delegated BigQuery Cloud resource connection service account rather than managing individual user keys.
In-place SQL inference: Once defined, users can invoke the remote model directly inside SQL queries using functions like ML.GENERATE_TEXT or ML.PREDICT.
Cross-region compatibility: Supports regional and multi-regional datasets pointing to Vertex AI regional endpoints.

Why This Is Optimal

View source documentation

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM `my_project.my_dataset.training_data`;

What This Option Proposes

This statement constructs an internal model training query using an AS SELECT query block with a TRANSFORM clause, attempting to pass the connection as a feature transformation parameter.

Why It Doesn't Meet the Requirement

Invokes model training instead of remote registration (✗): The AS SELECT ... clause is used for training internally trained models (such as Linear Regression, Logistic Regression, or K-means clustering). Remote models do not train on rows in BigQuery and do not take an AS SELECT training query.
Misuses the TRANSFORM clause (✗): The TRANSFORM clause in BigQuery ML is used to specify manual feature preprocessing functions (e.g., ML.STANDARD_SCALER, ML.BUCKETIZE) during training, not for specifying external connection parameters.
Missing REMOTE WITH CONNECTION (✗): Omits the mandatory keyword REMOTE WITH CONNECTION required to link the model resource to BigQuery's connection framework.

Key Characteristics of Internally Trained Models

Require an AS SELECT query to ingest training features and target labels from BigQuery tables.
Process bytes and consume BigQuery computing slots during model training.
Fully managed and executed within the BigQuery infrastructure rather than external API endpoints.

This Answer Would Be Correct If

You were training a built-in supervised BigQuery ML model (such as a logistic regression or boosted tree model) and applying custom feature preprocessing rules using SQL functions.

View source documentation

Last updated October 23, 2026

Registering Vertex AI Remote Models in BigQuery

Explanations

CREATE OR REPLACE EXTERNAL TABLE my_project.my_dataset.remote_llm WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

What This Option Proposes

Why It Doesn't Meet the Requirement

Key Characteristics of External Tables

This Answer Would Be Correct If

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm IMPORT FROM CONNECTION my_project.us.my_connection OPTIONS (MODEL_TYPE = 'VERTEX_AI');

What This Option Proposes

Why It Doesn't Meet the Requirement

Key Characteristics of Model Import vs. Remote Models

This Answer Would Be Correct If

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm REMOTE WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

What BigQuery Remote Model DDL Is

Remote Model DDL Anatomy

How It Meets the Requirements

Key Advantages and Capabilities

Why This Is Optimal

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM my_project.my_dataset.training_data;

What This Option Proposes

Why It Doesn't Meet the Requirement

Key Characteristics of Internally Trained Models

This Answer Would Be Correct If

Try 10 interactive questions like this one, free

Related questions

BigQuery Generative AI Functions

BigQuery Remote Connection Vertex AI Service Account

Cloud Data Fusion Wrangler

Cloud Storage Lifecycle matchesStorageClass Condition

Logs Explorer Query Editor

Cloud Code Debug on Kubernetes

Registering Vertex AI Remote Models in BigQuery

Explanations

CREATE OR REPLACE EXTERNAL TABLE my_project.my_dataset.remote_llm WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

What This Option Proposes

Why It Doesn't Meet the Requirement

Key Characteristics of External Tables

This Answer Would Be Correct If

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm IMPORT FROM CONNECTION my_project.us.my_connection OPTIONS (MODEL_TYPE = 'VERTEX_AI');

What This Option Proposes

Why It Doesn't Meet the Requirement

Key Characteristics of Model Import vs. Remote Models

This Answer Would Be Correct If

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm REMOTE WITH CONNECTION my_project.us.my_connection OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

What BigQuery Remote Model DDL Is

Remote Model DDL Anatomy

How It Meets the Requirements

Key Advantages and Capabilities

Why This Is Optimal

CREATE OR REPLACE MODEL my_project.my_dataset.remote_llm TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM my_project.my_dataset.training_data;

What This Option Proposes

Why It Doesn't Meet the Requirement

Key Characteristics of Internally Trained Models

This Answer Would Be Correct If

Try 10 interactive questions like this one, free

Related questions

BigQuery Generative AI Functions

BigQuery Remote Connection Vertex AI Service Account

Cloud Data Fusion Wrangler

Cloud Storage Lifecycle matchesStorageClass Condition

Logs Explorer Query Editor

Cloud Code Debug on Kubernetes

CREATE OR REPLACE EXTERNAL TABLE `my_project.my_dataset.remote_llm` WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` IMPORT FROM CONNECTION `my_project.us.my_connection` OPTIONS (MODEL_TYPE = 'VERTEX_AI');

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` REMOTE WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM `my_project.my_dataset.training_data`;

CREATE OR REPLACE EXTERNAL TABLE `my_project.my_dataset.remote_llm` WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/...');

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` IMPORT FROM CONNECTION `my_project.us.my_connection` OPTIONS (MODEL_TYPE = 'VERTEX_AI');

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` REMOTE WITH CONNECTION `my_project.us.my_connection` OPTIONS (ENDPOINT = 'https://us-central1-aiplatform.googleapis.com/v1/projects/my_project/locations/us-central1/publishers/google/models/text-bison');

CREATE OR REPLACE MODEL `my_project.my_dataset.remote_llm` TRANSFORM (REMOTE_CONNECTION = 'my_project.us.my_connection') AS SELECT * FROM `my_project.my_dataset.training_data`;