Skip to main content
Embeddings convert text and other content into numerical vectors that capture semantic meaning. These vectors can be used for similarity search, clustering, classification, and other machine learning tasks.

​
Generate embeddings

Generate embeddings for a single piece of content: 
response = client.models.embed_content(
    model='gemini-embedding-001',
    contents='why is the sky blue?',
)
print(response)

​
Batch embeddings

Generate embeddings for multiple pieces of content: 
from google.genai import types

response = client.models.embed_content(
    model='gemini-embedding-001',
    contents=['why is the sky blue?', 'What is your age?'],
    config=types.EmbedContentConfig(output_dimensionality=10),
)

print(response)

​
Dimensionality reduction

Reduce the output dimensionality of embeddings for efficiency: 
from google.genai import types

response = client.models.embed_content(
    model='gemini-embedding-001',
    contents='why is the sky blue?',
    config=types.EmbedContentConfig(output_dimensionality=256),
)

print(len(response.embeddings[0].values))  # 256
The output_dimensionality parameter allows you to specify a smaller dimension for the embedding vectors. This can reduce storage and computation costs while maintaining semantic meaning.

​
Embedding models

Available embedding models:
  • gemini-embedding-001 - General-purpose text embeddings
  • text-embedding-004 - Latest text embedding model with improved performance

​
Use cases

Find similar documents by comparing embedding vectors: 
import numpy as np

# Embed documents
docs = ['Document 1', 'Document 2', 'Document 3']
response = client.models.embed_content(
    model='gemini-embedding-001',
    contents=docs,
)
doc_embeddings = [emb.values for emb in response.embeddings]

# Embed query
query_response = client.models.embed_content(
    model='gemini-embedding-001',
    contents='search query',
)
query_embedding = query_response.embeddings[0].values

# Calculate cosine similarity
similarities = [
    np.dot(query_embedding, doc_emb) / 
    (np.linalg.norm(query_embedding) * np.linalg.norm(doc_emb))
    for doc_emb in doc_embeddings
]

# Find most similar document
most_similar_idx = np.argmax(similarities)
print(f"Most similar: {docs[most_similar_idx]}")

​
Clustering

Group similar content together using embedding vectors: 
from sklearn.cluster import KMeans

# Generate embeddings for your content
response = client.models.embed_content(
    model='gemini-embedding-001',
    contents=['text1', 'text2', 'text3', 'text4'],
)

embeddings = [emb.values for emb in response.embeddings]

# Cluster embeddings
kmeans = KMeans(n_clusters=2)
kmeans.fit(embeddings)

print(kmeans.labels_)

​
Classification

Use embeddings as features for classification tasks: 
from sklearn.linear_model import LogisticRegression

# Generate embeddings for training data
X_train_embeddings = [...]  # Your training embeddings
y_train = [...]  # Your labels

# Train classifier
clf = LogisticRegression()
clf.fit(X_train_embeddings, y_train)

# Predict on new content
new_response = client.models.embed_content(
    model='gemini-embedding-001',
    contents='new text',
)
new_embedding = [new_response.embeddings[0].values]
prediction = clf.predict(new_embedding)

​
Response format

The embed_content response contains:
  • embeddings - List of embedding objects
    • values - The embedding vector (list of floats)
    • content - The original content that was embedded

Build docs developers (and LLMs) love

Get started for freeTalk to us