Voyage AI

What is Voyage AI?

Voyage AI provides embedding and reranking models that help applications search and retrieve information from unstructured data. It is owned by MongoDB and is aimed at teams building retrieval-augmented generation (RAG) systems.

Key capabilities of Voyage AI

• Embedding models: Convert text into vectors for semantic search
• Rerankers: Re-rank retrieved results to improve relevance
• Long context: Context window up to 32K tokens
• Compact vectors: Vectors the vendor says are 3x to 8x shorter than competitors
• Domain models: Specialized models such as a legal embedding model
• Model-agnostic: Works with any vector database and LLM
• Flexible deployment: API, cloud marketplaces and custom on-premises

How Voyage AI works

You send text to the Voyage API and receive embedding vectors, which you store in any vector database. At query time you embed the question, retrieve candidates and optionally pass them through a reranker to sort by relevance before handing context to an LLM. There is no human review step in the loop.

Who uses Voyage AI?

Developers building RAG and search applications. The vendor cites Harvey, which reported a 25% drop in irrelevant retrieval and a threefold cut in vector database costs after adopting a legal embedding model.

Voyage AI pricing

Voyage AI uses consumption-based pricing through its own service and cloud marketplaces such as AWS. Per-token rates are not stated on the homepage, so check the vendor pricing page.

Voyage AI alternatives

Alternatives include Cohere for embeddings and rerank APIs, Mixedbread for open embedding and reranking models and Mindee for document data extraction.