Skip to main content
Voyage AI provides a range of text embedding models optimized for different use cases, from general-purpose retrieval to specialized domains like code, finance, and law.

​
Available models

You can use any of the following text embedding models with the Voyage AI provider:
ModelContext lengthEmbedding dimensionsModel ID
Voyage 3.532,000 tokens256, 512, 1024 (default), 2048voyage-3.5
Voyage 3.5 Lite32,000 tokens256, 512, 1024 (default), 2048voyage-3.5-lite
Voyage 3 Large32,000 tokens256, 512, 1024 (default), 2048voyage-3-large
Voyage 332,000 tokens1024voyage-3
Voyage 3 Lite32,000 tokens512voyage-3-lite
Voyage Code 332,000 tokens256, 512, 1024 (default), 2048voyage-code-3
Voyage Finance 232,000 tokens1024voyage-finance-2
Voyage Multilingual 232,000 tokens1024voyage-multilingual-2
Voyage Law 216,000 tokens1024voyage-law-2
Voyage Code 216,000 tokens1536voyage-code-2
Older models (voyage-large-2-instruct, voyage-large-2, voyage-2, voyage-02, voyage-01, voyage-lite-01) are deprecated and will be removed in the future. Use the latest models instead.

​
Usage example

You can create a text embedding model using the voyage.textEmbeddingModel() method: 
import { voyage } from 'voyage-ai-provider';
import { embedMany } from 'ai';

const embeddingModel = voyage.textEmbeddingModel('voyage-3-lite');

const { embeddings } = await embedMany({
  model: embeddingModel,
  values: [
    'Implement a binary search algorithm',
    'Create a REST API endpoint',
  ],
});

​
Model settings

You can customize model behavior by passing settings as the second argument: 
const embeddingModel = voyage.textEmbeddingModel('voyage-3-lite', {
  inputType: 'document',
  outputDimension: 1024,
  outputDtype: 'float',
  truncation: true,
});

​
Available settings

​
inputType
'query' | 'document'
default:"query"
The input type for the embeddings. Use query for search queries and document for documents being indexed.
  • For query: The prompt “Represent the query for retrieving supporting documents: ” is prepended
  • For document: The prompt “Represent the document for retrieval: ” is prepended
​
outputDimension
number
The number of dimensions for the resulting output embeddings. If not specified, uses the model’s default dimension.Models like voyage-3.5, voyage-3.5-lite, voyage-3-large, and voyage-code-3 support: 256, 512, 1024 (default), and 2048.
​
outputDtype
'float' | 'int8' | 'uint8' | 'binary' | 'ubinary'
default:"float"
The data type for the resulting output embeddings.
  • float: 32-bit single-precision floating-point numbers (supported by all models)
  • int8: 8-bit integers ranging from -128 to 127
  • uint8: 8-bit integers ranging from 0 to 255
  • binary: Bit-packed quantized single-bit values using int8
  • ubinary: Bit-packed quantized single-bit values using uint8
The int8, uint8, binary, and ubinary types are currently only supported by voyage-code-3.
​
truncation
boolean
Whether to truncate the input texts to fit within the context length. When enabled, long inputs are automatically truncated to the model’s maximum context length.

​
Choosing the right model

​
General-purpose tasks

  • voyage-3.5: Best overall performance for general retrieval tasks with flexible dimensions
  • voyage-3.5-lite: Faster inference with similar quality to voyage-3.5
  • voyage-3-large: High-quality embeddings for demanding retrieval applications
  • voyage-3: Balanced performance for standard use cases
  • voyage-3-lite: Lightweight option for cost-sensitive applications

​
Specialized domains

  • voyage-code-3: Optimized for code search, code similarity, and technical documentation
  • voyage-finance-2: Fine-tuned for financial documents and terminology
  • voyage-multilingual-2: Supports multiple languages for cross-lingual retrieval
  • voyage-law-2: Specialized for legal documents and terminology
  • voyage-code-2: Earlier code-specialized model (consider upgrading to voyage-code-3)
For code-related tasks, use voyage-code-3 with the appropriate outputDimension based on your accuracy vs. speed requirements. Higher dimensions generally provide better accuracy at the cost of storage and computation.

​
Advanced features

​
Variable output dimensions

For models that support multiple output dimensions (voyage-3.5, voyage-3.5-lite, voyage-3-large, voyage-code-3), you can trade off between accuracy and efficiency: 
// Higher accuracy, larger storage
const highAccuracy = voyage.textEmbeddingModel('voyage-code-3', {
  outputDimension: 2048,
});

// Balanced
const balanced = voyage.textEmbeddingModel('voyage-code-3', {
  outputDimension: 1024,
});

// Faster, smaller storage
const efficient = voyage.textEmbeddingModel('voyage-code-3', {
  outputDimension: 256,
});

​
Quantization

For applications requiring extreme efficiency, you can use quantized embeddings with voyage-code-3: 
const quantizedModel = voyage.textEmbeddingModel('voyage-code-3', {
  outputDtype: 'int8', // 4x smaller than float
  outputDimension: 1024,
});
Quantization reduces storage and memory requirements but may slightly impact retrieval accuracy. Test your specific use case to ensure acceptable performance.

Build docs developers (and LLMs) love

Get started for freeTalk to us