Tutorial da API Text-Embedding-3-Small - Guia do Modelo de Embeddings da OpenAI

Construindo um motor de busca semântica ou um sistema RAG (Retrieval-Augmented Generation)? Text-embedding-3-small é o mais recente modelo de embeddings da OpenAI que converte texto em vetores numéricos, possibilitando buscas por similaridade e recuperação de conteúdo extremamente eficientes.

Neste guia, você vai aprender:

O que são embeddings de texto e por que são importantes
Como usar a API text-embedding-3-small
Exemplos de código completos em Python e Node.js
Dimensões personalizadas para armazenamento otimizado
Comparação de preços e otimização de custos

What is Text-Embedding-3-Small?#

Text-embedding-3-small é o modelo compacto de embeddings da OpenAI lançado em janeiro de 2024. Ele converte texto em vetores de 1536 dimensões que capturam significado semântico, permitindo:

Busca Semântica: Encontrar documentos relevantes com base no significado, não apenas em palavras-chave
Sistemas RAG: Recuperar contexto para respostas de LLM
Correspondência de Similaridade: Comparar similaridade de textos para recomendações
Agrupamento (Clustering): Agrupar documentos semelhantes
Classificação: Categorizar texto com base no conteúdo

Model Specifications#

Specification	Value
Model Name	`text-embedding-3-small`
Default Dimensions	1536
Custom Dimensions	256, 512, 1024, 1536
Max Input Tokens	8,191
Output	Normalized vector

Quick Start#

Prerequisites#

Sign up at Crazyrouter
Get your API key from the dashboard
Python 3.8+ or Node.js 16+

Python Example#

python

from openai import OpenAI

client = OpenAI(
    api_key="your-crazyrouter-api-key",
    base_url="https://crazyrouter.com/v1"
)

# Generate embedding for a single text
response = client.embeddings.create(
    model="text-embedding-3-small",
    input="Machine learning is transforming industries worldwide."
)

embedding = response.data[0].embedding
print(f"Dimensions: {len(embedding)}")  # Output: 1536
print(f"First 5 values: {embedding[:5]}")

Node.js Example#

javascript

import OpenAI from 'openai';

const client = new OpenAI({
    apiKey: 'your-crazyrouter-api-key',
    baseURL: 'https://crazyrouter.com/v1'
});

async function getEmbedding(text) {
    const response = await client.embeddings.create({
        model: 'text-embedding-3-small',
        input: text
    });

    return response.data[0].embedding;
}

// Usage
const embedding = await getEmbedding('Machine learning is amazing');
console.log(`Dimensions: ${embedding.length}`);  // Output: 1536

cURL Example#

bash

curl -X POST https://crazyrouter.com/v1/embeddings \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "text-embedding-3-small",
    "input": "Hello world"
  }'

Response:

json

{
  "object": "list",
  "model": "text-embedding-3-small",
  "usage": {
    "prompt_tokens": 2,
    "total_tokens": 2
  },
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [-0.0020785425, -0.049085874, 0.02094679, ...]
    }
  ]
}

Batch Embedding#

Processe vários textos em uma única chamada de API para melhor eficiência:

python

from openai import OpenAI

client = OpenAI(
    api_key="your-crazyrouter-api-key",
    base_url="https://crazyrouter.com/v1"
)

# Batch embedding - multiple texts at once
texts = [
    "Python is a programming language",
    "JavaScript runs in browsers",
    "Machine learning uses neural networks"
]

response = client.embeddings.create(
    model="text-embedding-3-small",
    input=texts
)

# Access each embedding
for i, data in enumerate(response.data):
    print(f"Text {i}: {len(data.embedding)} dimensions")

# Output:
# Text 0: 1536 dimensions
# Text 1: 1536 dimensions
# Text 2: 1536 dimensions

Custom Dimensions#

Reduza os custos de armazenamento usando dimensões menores. O modelo suporta redução de dimensão mantendo a qualidade:

python

# Use 512 dimensions instead of 1536
response = client.embeddings.create(
    model="text-embedding-3-small",
    input="Your text here",
    dimensions=512  # Options: 256, 512, 1024, 1536
)

embedding = response.data[0].embedding
print(f"Dimensions: {len(embedding)}")  # Output: 512

Dimension Comparison#

Dimensions	Storage (per vector)	Use Case
256	1 KB	Aplicativos móveis, armazenamento limitado
512	2 KB	Desempenho equilibrado
1024	4 KB	Necessidades de alta precisão
1536	6 KB	Máxima precisão

Building a Semantic Search System#

Aqui está um exemplo completo de construção de um sistema de busca semântica:

python

import numpy as np
from openai import OpenAI

client = OpenAI(
    api_key="your-crazyrouter-api-key",
    base_url="https://crazyrouter.com/v1"
)

def get_embedding(text):
    """Get embedding for a single text"""
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=text
    )
    return response.data[0].embedding

def cosine_similarity(a, b):
    """Calculate cosine similarity between two vectors"""
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

# Document database
documents = [
    "Python is great for data science and machine learning",
    "JavaScript is essential for web development",
    "Docker containers simplify deployment",
    "Kubernetes orchestrates container workloads",
    "PostgreSQL is a powerful relational database"
]

# Pre-compute embeddings for all documents
doc_embeddings = [get_embedding(doc) for doc in documents]

# Search function
def search(query, top_k=3):
    query_embedding = get_embedding(query)

    # Calculate similarities
    similarities = [
        cosine_similarity(query_embedding, doc_emb)
        for doc_emb in doc_embeddings
    ]

    # Get top results
    results = sorted(
        zip(documents, similarities),
        key=lambda x: x[1],
        reverse=True
    )[:top_k]

    return results

# Example search
results = search("How to deploy applications?")
for doc, score in results:
    print(f"Score: {score:.4f} - {doc}")

# Output:
# Score: 0.8234 - Docker containers simplify deployment
# Score: 0.7891 - Kubernetes orchestrates container workloads
# Score: 0.6543 - PostgreSQL is a powerful relational database

Integration with Vector Databases#

Pinecone Integration#

python

import pinecone
from openai import OpenAI

# Initialize clients
client = OpenAI(
    api_key="your-crazyrouter-api-key",
    base_url="https://crazyrouter.com/v1"
)

pinecone.init(api_key="your-pinecone-key")
index = pinecone.Index("your-index")

def embed_and_upsert(texts, ids):
    """Embed texts and store in Pinecone"""
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=texts
    )

    vectors = [
        (id, data.embedding)
        for id, data in zip(ids, response.data)
    ]

    index.upsert(vectors=vectors)

def search_pinecone(query, top_k=5):
    """Search Pinecone with query embedding"""
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=query
    )

    results = index.query(
        vector=response.data[0].embedding,
        top_k=top_k
    )

    return results

ChromaDB Integration#

python

import chromadb
from openai import OpenAI

client = OpenAI(
    api_key="your-crazyrouter-api-key",
    base_url="https://crazyrouter.com/v1"
)

# Initialize ChromaDB
chroma_client = chromadb.Client()
collection = chroma_client.create_collection("documents")

def get_embeddings(texts):
    """Get embeddings for multiple texts"""
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=texts
    )
    return [data.embedding for data in response.data]

# Add documents
documents = ["doc1 content", "doc2 content", "doc3 content"]
embeddings = get_embeddings(documents)

collection.add(
    embeddings=embeddings,
    documents=documents,
    ids=["doc1", "doc2", "doc3"]
)

# Query
query_embedding = get_embeddings(["search query"])[0]
results = collection.query(
    query_embeddings=[query_embedding],
    n_results=3
)

Available Embedding Models#

Crazyrouter fornece acesso a vários modelos de embeddings da OpenAI:

Model	Dimensions	Price Ratio	Best For
`text-embedding-3-small`	1536	0.01	Uso geral, melhor custo-benefício
`text-embedding-3-large`	3072	0.065	Necessidades de alta precisão
`text-embedding-ada-002`	1536	0.05	Compatibilidade legada

Pricing Comparison#

Provider	Model	Price per 1M tokens
OpenAI Official	text-embedding-3-small	$0.020
Crazyrouter	text-embedding-3-small	$0.002
OpenAI Official	text-embedding-3-large	$0.130
Crazyrouter	text-embedding-3-large	$0.013

Pricing Disclaimer: Prices shown are for demonstration and may change. Actual billing is based on real-time prices at request time.

Cost Savings Example:

Para um sistema RAG processando 10M tokens/mês:

OpenAI Official: $200/mês
Crazyrouter: $20/mês
Economia: 90%

Best Practices#

1. Batch Your Requests#

python

# Good - single API call for multiple texts
response = client.embeddings.create(
    model="text-embedding-3-small",
    input=["text1", "text2", "text3"]  # Up to 2048 texts
)

# Bad - multiple API calls
for text in texts:
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=text
    )

2. Cache Embeddings#

python

import hashlib
import json

embedding_cache = {}

def get_embedding_cached(text):
    # Create cache key
    cache_key = hashlib.md5(text.encode()).hexdigest()

    if cache_key in embedding_cache:
        return embedding_cache[cache_key]

    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=text
    )

    embedding = response.data[0].embedding
    embedding_cache[cache_key] = embedding

    return embedding

3. Use Appropriate Dimensions#

256 dimensions: Aplicativos móveis, dispositivos IoT
512 dimensions: Aplicações web com limitações de armazenamento
1024 dimensions: Aplicações padrão
1536 dimensions: Requisitos de máxima precisão

Frequently Asked Questions#

What's the difference between text-embedding-3-small and text-embedding-3-large?#

Text-embedding-3-small produz vetores de 1536 dimensões e é otimizado para custo-benefício. Text-embedding-3-large produz vetores de 3072 dimensões com maior precisão, porém com custo 6,5x maior. Para a maioria das aplicações, text-embedding-3-small oferece excelentes resultados.

Can I reduce dimensions after generating embeddings?#

Sim, você pode usar o parâmetro dimensions para gerar vetores menores diretamente. Isso é mais eficiente do que gerar vetores completos e depois truncá-los.

How many texts can I embed in one request?#

Você pode gerar embeddings para até 2048 textos em uma única requisição de API. Para grandes conjuntos de dados, faça o processamento em lotes de 2048.

Are the embeddings normalized?#

Sim, text-embedding-3-small retorna vetores normalizados (comprimento unitário), então você pode usar produto escalar em vez de similaridade de cosseno para computação mais rápida.

Getting Started#

Sign up at Crazyrouter
Get your API key from the dashboard
Install the SDK: pip install openai or npm install openai
Start embedding with the code examples above

Related Articles:

For questions, contact support@crazyrouter.com

Tutorial da API Text-Embedding-3-Small - Guia do Modelo de Embeddings da OpenAI

What is Text-Embedding-3-Small?#

Model Specifications#

Quick Start#

Prerequisites#

Python Example#

Node.js Example#

cURL Example#

Batch Embedding#

Custom Dimensions#

Dimension Comparison#

Building a Semantic Search System#

Integration with Vector Databases#

Pinecone Integration#

ChromaDB Integration#

Available Embedding Models#

Pricing Comparison#

Best Practices#

1. Batch Your Requests#

2. Cache Embeddings#

3. Use Appropriate Dimensions#

Frequently Asked Questions#

What's the difference between text-embedding-3-small and text-embedding-3-large?#

Can I reduce dimensions after generating embeddings?#

How many texts can I embed in one request?#

Are the embeddings normalized?#

Getting Started#

Related Articles

Guia de Instalação e Uso do Claude Code - Configuração do Assistente de Programação com IA

Doubao Seed Code: Modelo de Geração de Código com IA da ByteDance - Guia Completo da API

Melhores Geradores de Música com IA 2026: Suno vs Udio vs Stable Audio Comparados