SQL Keywords

AI_EMBED

Feb 23, 2026

min read

AI_EMBED

Overview

Generates embedding vectors for text or images, which can be used for similarity search, clustering, classification, and semantic search tasks.

Syntax

AI_EMBED(
  model_name,
  input
)

AI_EMBED(
  model_name,
  input
)

AI_EMBED(
  model_name,
  input
)

Parameters

model_name (VARCHAR): Embedding model to use (e.g., 'snowflake-arctic-embed-l-v2.0', 'multilingual-e5-large')
input (VARCHAR or FILE): Text string or image file to embed

Use Cases

Semantic search and similarity matching
Document clustering and grouping
Recommendation systems
Duplicate detection
Content-based filtering
Multi-modal search (text + images)

Available Models

Text Embedding Models (768 dimensions)

e5-base-v2 - Context window: 512 tokens
snowflake-arctic-embed-m - Context window: 512 tokens
snowflake-arctic-embed-m-v1.5 - Context window: 512 tokens

Text Embedding Models (1024 dimensions)

snowflake-arctic-embed-l-v2.0 - Context window: 128K tokens
snowflake-arctic-embed-l-v2.0-8k - Context window: 8K tokens
nv-embed-qa-4 - Context window: 512 tokens
multilingual-e5-large - Context window: 512 tokens
voyage-multilingual-2 - Context window: 32K tokens

Code Examples

Example 1: Generate Text Embeddings

SELECT 
    product_id,
    product_description,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        product_description
    ) AS product_embedding
FROM

SELECT 
    product_id,
    product_description,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        product_description
    ) AS product_embedding
FROM

SELECT 
    product_id,
    product_description,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        product_description
    ) AS product_embedding
FROM

Example 2: Find Similar Documents

WITH document_embeddings AS (
    SELECT 
        doc_id,
        doc_text,
        AI_EMBED('multilingual-e5-large', doc_text) AS embedding
    FROM documents
),
query_embedding AS (
    SELECT AI_EMBED(
        'multilingual-e5-large',
        'machine learning best practices'
    ) AS query_vec
)
SELECT 
    d.doc_id,
    d.doc_text,
    VECTOR_COSINE_SIMILARITY(d.embedding, q.query_vec) AS similarity_score
FROM document_embeddings d, query_embedding q
ORDER BY similarity_score DESC
LIMIT 10

WITH document_embeddings AS (
    SELECT 
        doc_id,
        doc_text,
        AI_EMBED('multilingual-e5-large', doc_text) AS embedding
    FROM documents
),
query_embedding AS (
    SELECT AI_EMBED(
        'multilingual-e5-large',
        'machine learning best practices'
    ) AS query_vec
)
SELECT 
    d.doc_id,
    d.doc_text,
    VECTOR_COSINE_SIMILARITY(d.embedding, q.query_vec) AS similarity_score
FROM document_embeddings d, query_embedding q
ORDER BY similarity_score DESC
LIMIT 10

WITH document_embeddings AS (
    SELECT 
        doc_id,
        doc_text,
        AI_EMBED('multilingual-e5-large', doc_text) AS embedding
    FROM documents
),
query_embedding AS (
    SELECT AI_EMBED(
        'multilingual-e5-large',
        'machine learning best practices'
    ) AS query_vec
)
SELECT 
    d.doc_id,
    d.doc_text,
    VECTOR_COSINE_SIMILARITY(d.embedding, q.query_vec) AS similarity_score
FROM document_embeddings d, query_embedding q
ORDER BY similarity_score DESC
LIMIT 10

Example 3: Create Embedding Table for Vector Search

CREATE TABLE product_embeddings AS
SELECT 
    product_id,
    product_name,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        product_name || ' ' || description
    ) AS embedding_vector
FROM

CREATE TABLE product_embeddings AS
SELECT 
    product_id,
    product_name,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        product_name || ' ' || description
    ) AS embedding_vector
FROM

CREATE TABLE product_embeddings AS
SELECT 
    product_id,
    product_name,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        product_name || ' ' || description
    ) AS embedding_vector
FROM

Example 4: Multilingual Embeddings

SELECT 
    article_id,
    language,
    AI_EMBED(
        'multilingual-e5-large',
        article_content
    ) AS content_embedding
FROM international_articles
WHERE language IN ('en', 'es', 'fr', 'de')

SELECT 
    article_id,
    language,
    AI_EMBED(
        'multilingual-e5-large',
        article_content
    ) AS content_embedding
FROM international_articles
WHERE language IN ('en', 'es', 'fr', 'de')

SELECT 
    article_id,
    language,
    AI_EMBED(
        'multilingual-e5-large',
        article_content
    ) AS content_embedding
FROM international_articles
WHERE language IN ('en', 'es', 'fr', 'de')

Example 5: Image Embeddings

SELECT 
    image_id,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        TO_FILE('@images/' || image_filename)
    ) AS image_embedding
FROM

SELECT 
    image_id,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        TO_FILE('@images/' || image_filename)
    ) AS image_embedding
FROM

SELECT 
    image_id,
    AI_EMBED(
        'snowflake-arctic-embed-l-v2.0',
        TO_FILE('@images/' || image_filename)
    ) AS image_embedding
FROM

Data Output Examples

Text Embedding Output

Input: "Snowflake is a cloud data platform"
Model: snowflake-arctic-embed-l-v2.0

Output: [0.023, -0.156, 0.089, ..., 0.234]  // 1024-dimensional vector

Input: "Snowflake is a cloud data platform"
Model: snowflake-arctic-embed-l-v2.0

Output: [0.023, -0.156, 0.089, ..., 0.234]  // 1024-dimensional vector

Input: "Snowflake is a cloud data platform"
Model: snowflake-arctic-embed-l-v2.0

Output: [0.023, -0.156, 0.089, ..., 0.234]  // 1024-dimensional vector

Similarity Search Result

Query: "How to optimize SQL queries?"

Top Matches:
1. "SQL Query Performance Tuning" - similarity: 0.89
2. "Database Optimization Techniques" - similarity: 0.85
3. "Best Practices for Query Efficiency" - similarity: 0.82

Query: "How to optimize SQL queries?"

Top Matches:
1. "SQL Query Performance Tuning" - similarity: 0.89
2. "Database Optimization Techniques" - similarity: 0.85
3. "Best Practices for Query Efficiency" - similarity: 0.82

Query: "How to optimize SQL queries?"

Top Matches:
1. "SQL Query Performance Tuning" - similarity: 0.89
2. "Database Optimization Techniques" - similarity: 0.85
3. "Best Practices for Query Efficiency" - similarity: 0.82

Limitations & Considerations

Token Limits

Each model has specific context window limits
Exceeding limits causes truncation or errors
Use AI_COUNT_TOKENS to verify input size

Dimension Sizes

768-dimensional models: Faster, less storage
1024-dimensional models: More accurate, larger storage
Choose based on accuracy vs. performance needs

Cost

Billing based on input tokens only
No output token charges
Embedding generation is relatively inexpensive

Regional Availability

AWS US West/East: ✓
Azure East US: ✓
EU regions: ✓
Cross-region inference: ✓

Best Practices

1. Choose the Right Model

-- For long documents
AI_EMBED('snowflake-arctic-embed-l-v2.0', long_text)  -- 128K context

-- For short texts (queries, titles)
AI_EMBED('snowflake-arctic-embed-m', short_text)  -- 512 tokens, faster

-- For multilingual content
AI_EMBED('multilingual-e5-large', text)  -- Supports 100+ languages

-- For long documents
AI_EMBED('snowflake-arctic-embed-l-v2.0', long_text)  -- 128K context

-- For short texts (queries, titles)
AI_EMBED('snowflake-arctic-embed-m', short_text)  -- 512 tokens, faster

-- For multilingual content
AI_EMBED('multilingual-e5-large', text)  -- Supports 100+ languages

-- For long documents
AI_EMBED('snowflake-arctic-embed-l-v2.0', long_text)  -- 128K context

-- For short texts (queries, titles)
AI_EMBED('snowflake-arctic-embed-m', short_text)  -- 512 tokens, faster

-- For multilingual content
AI_EMBED('multilingual-e5-large', text)  -- Supports 100+ languages

2. Store Embeddings for Reuse

-- Create materialized embedding table
CREATE TABLE cached_embeddings AS
SELECT 
    id,
    content,
    AI_EMBED('snowflake-arctic-embed-l-v2.0', content) AS embedding,
    CURRENT_TIMESTAMP() AS embedding_created_at
FROM source_data;

-- Add index for faster vector search
CREATE INDEX embedding_idx ON cached_embeddings(embedding)

-- Create materialized embedding table
CREATE TABLE cached_embeddings AS
SELECT 
    id,
    content,
    AI_EMBED('snowflake-arctic-embed-l-v2.0', content) AS embedding,
    CURRENT_TIMESTAMP() AS embedding_created_at
FROM source_data;

-- Add index for faster vector search
CREATE INDEX embedding_idx ON cached_embeddings(embedding)

-- Create materialized embedding table
CREATE TABLE cached_embeddings AS
SELECT 
    id,
    content,
    AI_EMBED('snowflake-arctic-embed-l-v2.0', content) AS embedding,
    CURRENT_TIMESTAMP() AS embedding_created_at
FROM source_data;

-- Add index for faster vector search
CREATE INDEX embedding_idx ON cached_embeddings(embedding)

3. Normalize Input Text

-- Clean and prepare text before embedding
SELECT AI_EMBED(
    'snowflake-arctic-embed-l-v2.0',
    LOWER(TRIM(REGEXP_REPLACE(text, '[^a-zA-Z0-9\s]', '')))
) AS embedding
FROM

-- Clean and prepare text before embedding
SELECT AI_EMBED(
    'snowflake-arctic-embed-l-v2.0',
    LOWER(TRIM(REGEXP_REPLACE(text, '[^a-zA-Z0-9\s]', '')))
) AS embedding
FROM

-- Clean and prepare text before embedding
SELECT AI_EMBED(
    'snowflake-arctic-embed-l-v2.0',
    LOWER(TRIM(REGEXP_REPLACE(text, '[^a-zA-Z0-9\s]', '')))
) AS embedding
FROM

4. Batch Processing

-- Process embeddings in batches for large datasets
CREATE OR REPLACE TABLE product_embeddings AS
SELECT 
    product_id,
    AI_EMBED('snowflake-arctic-embed-l-v2.0', description) AS embedding
FROM products
WHERE embedding_version IS NULL
LIMIT 10000;  -- Process in chunks

-- Process embeddings in batches for large datasets
CREATE OR REPLACE TABLE product_embeddings AS
SELECT 
    product_id,
    AI_EMBED('snowflake-arctic-embed-l-v2.0', description) AS embedding
FROM products
WHERE embedding_version IS NULL
LIMIT 10000;  -- Process in chunks

-- Process embeddings in batches for large datasets
CREATE OR REPLACE TABLE product_embeddings AS
SELECT 
    product_id,
    AI_EMBED('snowflake-arctic-embed-l-v2.0', description) AS embedding
FROM products
WHERE embedding_version IS NULL
LIMIT 10000;  -- Process in chunks

Use with Cortex Search

-- Create Cortex Search Service with embeddings
CREATE CORTEX SEARCH SERVICE product_search
ON embedding_vector
ATTRIBUTES product_name, category
WAREHOUSE = compute_wh
TARGET_LAG = '1 hour'
AS (
    SELECT 
        product_id,
        product_name,
        category,
        AI_EMBED('snowflake-arctic-embed-l-v2.0', description) AS embedding_vector
    FROM products
)

-- Create Cortex Search Service with embeddings
CREATE CORTEX SEARCH SERVICE product_search
ON embedding_vector
ATTRIBUTES product_name, category
WAREHOUSE = compute_wh
TARGET_LAG = '1 hour'
AS (
    SELECT 
        product_id,
        product_name,
        category,
        AI_EMBED('snowflake-arctic-embed-l-v2.0', description) AS embedding_vector
    FROM products
)

-- Create Cortex Search Service with embeddings
CREATE CORTEX SEARCH SERVICE product_search
ON embedding_vector
ATTRIBUTES product_name, category
WAREHOUSE = compute_wh
TARGET_LAG = '1 hour'
AS (
    SELECT 
        product_id,
        product_name,
        category,
        AI_EMBED('snowflake-arctic-embed-l-v2.0', description) AS embedding_vector
    FROM products
)

Related Functions

AI_SIMILARITY - Calculate similarity between embeddings
VECTOR_COSINE_SIMILARITY - Compute cosine similarity
AI_COUNT_TOKENS - Check input token count

Interested to Learn More?
Try Out the Free 14-Days Trial

Start free trial

Learn

Feb 18, 2026

BigQuery Global Queries: How to Run Cross-Region SQL in 2026

Learn

Feb 18, 2026

BigQuery Global Queries: How to Run Cross-Region SQL in 2026

Learn

Feb 18, 2026

BigQuery Global Queries: How to Run Cross-Region SQL in 2026

Analytics

Feb 18, 2026

Context Engineering and AI Quality for Data Teams

Analytics

Feb 18, 2026

Context Engineering and AI Quality for Data Teams

Analytics

Feb 18, 2026

Context Engineering and AI Quality for Data Teams

Product

Feb 18, 2026

Accelerate Analytics Development with Paradime and Tableau

Product

Feb 18, 2026

Accelerate Analytics Development with Paradime and Tableau

Product

Feb 18, 2026

Accelerate Analytics Development with Paradime and Tableau

Experience Analytics for the AI-Era

Start your 14-day trial today - it's free and no credit card needed

Start for free

Experience Analytics for the AI-Era

Start your 14-day trial today - it's free and no credit card needed

Start for free

Experience Analytics for the AI-Era

Start your 14-day trial today - it's free and no credit card needed

Start for free

Platform

Resources

ADD-ONs

Industries

About

Legal

Made with ❤️ in San Francisco ・ London

*dbt® and dbt Core® are federally registered trademarks of dbt Labs, Inc. in the United States and various jurisdictions around the world. Paradime is not a partner of dbt Labs. All rights therein are reserved to dbt Labs. Paradime is not a product or service of or endorsed by dbt Labs, Inc.

Start for free

Platform

Resources