Text Embeddings Fresh 🌱

Generate high-quality text embeddings with configurable dimensions, task types, and multi-language support.

Overview

Specifications

Parameter	Value
Model	gemini-embedding-2-preview
Max input tokens	8,192
Default dimensions	3,072
Dimension range	128 - 3,072
Languages	100+
Pricing (Standard)	$0.20 / 1M tokens
Pricing (Batch)	$0.10 / 1M tokens

Step 1: Basic Text Embedding

Python

from google import genai

client = genai.Client()

result = client.models.embed_content(
    model='gemini-embedding-2-preview',
    contents='What is the meaning of life?'
)

embedding = result.embeddings[0].values
print(f"Dimensions: {len(embedding)}")  # 3072

Node.js

import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI({});
const response = await ai.models.embedContent({
    model: 'gemini-embedding-2-preview',
    contents: 'What is the meaning of life?',
});
console.log(response.embeddings[0].values.length);  // 3072

cURL

curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-embedding-2-preview:embedContent" \
    -H "Content-Type: application/json" \
    -H "x-goog-api-key: ${GEMINI_API_KEY}" \
    -d '{"content": {"parts": [{"text": "What is the meaning of life?"}]}}'

Step 2: Specify Task Type

Task types optimize the embedding for your specific use case.

from google import genai
from google.genai import types

client = genai.Client()

# For search queries
query_embedding = client.models.embed_content(
    model='gemini-embedding-2-preview',
    contents='How to train a neural network',
    config=types.EmbedContentConfig(
        task_type='RETRIEVAL_QUERY'
    )
)

# For documents being indexed
doc_embedding = client.models.embed_content(
    model='gemini-embedding-2-preview',
    contents='Neural networks are trained using backpropagation...',
    config=types.EmbedContentConfig(
        task_type='RETRIEVAL_DOCUMENT'
    )
)

Step 3: Custom Dimensions

Reduce dimensions to save storage while maintaining quality.

from google import genai
from google.genai import types

client = genai.Client()

# 768 dimensions — good balance of quality and size
result = client.models.embed_content(
    model='gemini-embedding-2-preview',
    contents='Your text here',
    config=types.EmbedContentConfig(
        output_dimensionality=768
    )
)

print(f"Dimensions: {len(result.embeddings[0].values)}")  # 768

Normalization Required

When using dimensions below 3072, normalize your vectors before computing cosine similarity for accurate results.

Step 4: Batch Multiple Texts

Generate separate embeddings for multiple texts in one request.

from google import genai

client = genai.Client()

texts = [
    'First document about machine learning',
    'Second document about natural language processing',
    'Third document about computer vision',
]

result = client.models.embed_content(
    model='gemini-embedding-2-preview',
    contents=texts
)

for i, emb in enumerate(result.embeddings):
    print(f"Text {i+1}: {len(emb.values)} dimensions")

Single vs Multiple

• Single content with multiple parts = one aggregated embedding
• Multiple entries in contents array = separate embeddings per entry

Verification Checklist

• [ ] Text embedding generated with default 3072 dimensions
• [ ] Task type set appropriately for use case
• [ ] Custom dimensions tested (768, 1536)
• [ ] Normalization applied when using reduced dimensions
• [ ] Batch embedding tested with multiple texts
• [ ] Multi-language text tested if applicable

Troubleshooting

Issue	Solution
Token limit exceeded	Split text into chunks under 8192 tokens
Low similarity scores	Use matching task types (QUERY + DOCUMENT)
Dimension mismatch	Ensure consistent output_dimensionality across all embeddings
Slow responses	Use batch API for 50% cost savings on large volumes

See Also

• Getting Started — Initial setup
• Multimodal Embeddings — Combine text with other modalities
• Benchmarks — MTEB scores by dimension
• RAG Pipeline — End-to-end retrieval workflow