Models Overview

Voyage AI provides state-of-the-art embedding and reranking models. MongoDB's Embedding and Reranking API provides access to the latest Voyage AI models. This page describes the available models and when to use them.

What are embedding models and rerankers?

Choosing a Model

Expand this section to learn more about which model to choose for your use case.

For text embeddings, we recommend:

voyage-4-large for the best quality
voyage-4-lite for the lowest latency and cost
voyage-4 for a balance between quality and performance
A domain-specific model if your application is in one of the listed domains.

For other use cases, we recommend:

voyage-multimodal-3.5 for text, image, and video embeddings
voyage-context-4 for chunk-level and document-level retrieval tasks
rerank-2.5 for adding reranking to most applications
rerank-2.5-lite for adding reranking to latency-sensitive applications

Text Embeddings

Voyage AI provides the following text embedding models to capture the semantic meaning of text.

For details and example usage, see Text Embeddings.

General-Purpose Models

Use the following models for most AI search and retrieval applications.

Model	Context Length	Dimensions	Description
`voyage-4-large`	32,000 tokens	1024 (default), 256, 512, 2048	The best general-purpose and multilingual retrieval quality. All embeddings created with the 4 series are compatible with each other. To learn more, see the blog post.
`voyage-4`	32,000 tokens	1024 (default), 256, 512, 2048	Optimized for general-purpose and multilingual retrieval quality. All embeddings created with the 4 series are compatible with each other. To learn more, see the blog post.
`voyage-4-lite`	32,000 tokens	1024 (default), 256, 512, 2048	Optimized for latency and cost. All embeddings created with the 4 series are compatible with each other. To learn more, see the blog post.

Domain-Specific Models

Use the following models for specialized domains to achieve better accuracy.

Model	Context Length	Dimensions	Description
`voyage-code-3`	32,000 tokens	1024 (default), 256, 512, 2048	Optimized for code retrieval and documentation. To learn more, see the blog post.
`voyage-finance-2`	32,000 tokens	1024	Optimized for finance retrieval and RAG applications. To learn more, see the blog post.
`voyage-law-2`	16,000 tokens	1024	Optimized for legal retrieval and RAG applications. To learn more, see the blog post.

Open Models

Voyage also provides the following open-weight models.

Model	Context Length	Dimensions	Description
`voyage-4-nano`	32,000 tokens	512 (default), 128, 256	Open-weight model available on Hugging Face. All embeddings created with the 4 series are compatible with eachother To learn more, see the blog post.

Older Text Embedding Models

The following older models are still accessible from the API, but we recommend using the new models for better quality and efficiency.

The latest models perform better than the legacy models in all aspects, such as quality, context length, latency, and throughput.

Model	Context Length	Dimensions	Description
`voyage-3-large`	32,000 tokens	1024 (default), 256, 512, 2048	Previous generation of text embeddings for general-purpose and multilingual retrieval quality. To learn more, see the blog post.
`voyage-3.5`	32,000 tokens	1024 (default), 256, 512, 2048	Previous generation of text embeddings optimized for general-purpose and multilingual retrieval quality. To learn more, see the blog post.
`voyage-3.5-lite`	32,000 tokens	1024 (default), 256, 512, 2048	Previous generation of text embeddings optimized for latency and cost. To learn more, see the blog post.
`voyage-code-2`	16,000 tokens	1536	Optimized for code retrieval (17% better than alternatives). Previous generation of code embeddings. To learn more, see the blog post.

Contextualized Chunk Embeddings

Voyage AI provides the following models that generate embeddings while incorporating surrounding context for improved retrieval accuracy.

For details and example usage, see Contextualized Chunk Embeddings.

Model	Context Length	Dimensions	Description
In preview: `voyage-context-4`	120,000 tokens	1024 (default), 256, 512, 2048	Contextualized chunk embeddings optimized for general-purpose and multilingual retrieval quality.
`voyage-context-3`	120,000 tokens	1024 (default), 256, 512, 2048	Contextualized chunk embeddings optimized for general-purpose and multilingual retrieval quality. To learn more, see the blog post.

Multimodal Embeddings

Voyage AI provides the following embedding models that process text, images, and video.

For details and example usage, see Multimodal Embeddings.

Model	Context Length	Dimensions	Description
`voyage-multimodal-3.5`	32,000 tokens	1024 (default), 256, 512, 2048	Rich multimodal embedding model that can vectorize interleaved text and visual data, such as screenshots of PDFs, slides, tables, figures, videos, and more. To learn more, see the blog post.

Older Multimodal Models

The following older models are still accessible from the API, but we recommend using the new models for better quality and efficiency.

The latest models perform better than the legacy models in all aspects, such as quality, context length, latency, and throughput.

Model	Context Length	Dimensions	Description
`voyage-multimodal-3`	32,000 tokens	1024	Processes text and images into unified embeddings. Supports images from 50,000 to 2 million pixels. To learn more, see the blog post.

Rerankers

Voyage AI provides the following reranking models to refine your search results.

For details and example usage, see Rerankers.

Model	Context Length	Description
`rerank-2.5`	32,000	Highest accuracy. Recommended for most applications. To learn more, see the blog post.
`rerank-2.5-lite`	32,000	Fast and cost-effective model optimized for latency-sensitive applications. To learn more, see the blog post.

Older Rerankers

The following older models are still accessible from the API, but we recommend using the new models for better quality and efficiency.

The latest models perform better than the legacy models in all aspects, such as quality, context length, latency, and throughput.

Model	Context Length	Description
`rerank-2`	16,000 tokens	Our generalist second-generation reranker optimized for quality with multilingual support. To learn more, see the blog post.
`rerank-2-lite`	8,000 tokens	Our generalist second-generation reranker optimized for both latency and quality with multilingual support. To learn more, see the blog post.

Pricing

Model pricing is usage-based, with charges billed to the Atlas account linked to the API key used for access. All models include a free tier. Get started with 200 million free tokens for most models, or 50 million tokens for specialized models.

Pricing is based on the number of tokens in your documents and queries. The free tier includes 200 million tokens for most models, and 50 million tokens for the following specialized models: voyage-finance-2, voyage-law-2, voyage-code-2.

Model	Price per 1K tokens	Price per 1M tokens	Free tokens
`voyage-4-large`	$0.00012	$0.12	200 million
`voyage-4`	$0.00006	$0.06	200 million
`voyage-4-lite`	$0.00002	$0.02	200 million
`voyage-code-3`	$0.00018	$0.18	200 million
`voyage-finance-2` `voyage-law-2` `voyage-code-2`	$0.00012	$0.12	50 million

Older Text Embedding Models

The following table shows the pricing for older text embedding models. Free tokens are not offered for these models.

Model	Price per 1K tokens	Price per 1M tokens
`voyage-3-large`	$0.00018	$0.18
`voyage-3.5`	$0.00006	$0.06
`voyage-3.5-lite`	$0.00002	$0.02

Pricing is based on the number of tokens in your documents and queries.

Model	Price per 1K tokens	Price per 1M tokens	Free tokens
`voyage-context-4`	$0.00012	$0.12	200 million

Pricing is based on text tokens and image pixels. The free tier includes 200 million text tokens and 150 billion pixels for multimodal models. Images are processed between 50,000 pixels (minimum) and 2 million pixels (maximum), with costs ranging from $0.00003 to $0.0012 per image. For pricing purposes, each video frame is considered an image.

Note

Images with fewer than 50,000 pixels are upscaled, processed, and charged as a 50,000-pixel image. Images containing over 2 million pixels are downsampled and charged as 2 million-pixel images.

Model	Price per 1M tokens	Price per 1B pixels	Free tier
`voyage-multimodal-3.5`	$0.12	$0.60	200M tokens, 150B pixels

Image resolution	Number of pixels	Price per image	Price per 1K images
200px × 200px	40,000	$0.00003	$0.03
1000px × 1000px	1 million	$0.0006	$0.60
2000px × 2000px	4 million	$0.0012	$1.20
4000px × 4000px	16 million	$0.0012	$1.20

Example

The cost to vectorize a single input with 1,000 text tokens ($0.00012) and two 4 million-pixel images (2 × $0.0012) would be $0.00252.

Older Multimodal Models

The following table shows the pricing for older multimodal models. Free tokens are not offered for these models.

Model	Price per 1M tokens	Price per 1B pixels	Free tier
`voyage-multimodal-3`	$0.12	$0.60	0

Pricing is based on total processed tokens, calculated as (query tokens × number of documents) + sum of tokens in all documents. The free tier includes 200 million tokens for the latest reranker models.

Model	Price per 1K tokens	Price per 1M tokens	Est. price per request*	Free tokens
`rerank-2.5`	$0.00005	$0.05	$0.0025	200 million
`rerank-2.5-lite`	$0.00002	$0.02	$0.001	200 million

* Estimated price assumes 100 documents per request, with the sum of query tokens and tokens per document totaling 500.

Older Rerankers

The following table shows the pricing for older rerankers. Free tokens are not offered for these models.

Model	Price per 1K tokens	Price per 1M tokens	Est. price per request*	Free tokens
`rerank-2`	$0.00005	$0.05	$0.0025	0
`rerank-2-lite`	$0.00002	$0.02	$0.001	0

Back

Quick Start

API and Clients