Voyage AI by MongoDB

Supercharging Search and Retrieval for Unstructured Data

Build relevant, reliable AI applications with high-precision Voyage AI embeddings and rerankers in the Intelligent Data Platform from MongoDB.

Get Started

Embeddings and Rerankers

How embeddings and rerankers power RAG

The pipeline behind higher-quality RAG responses, from unstructured data to factual answers at lower cost.

On RTEB, the benchmark built to reflect real enterprise retrieval instead of academic datasets, Voyage AI models consistently place among the top performers.

A query flows into the Voyage AI and MongoDB platform — embedding models, unstructured data, rerankers and vector search — then to an LLM, producing a contextually rich, grounded response.

Powered by Cutting-Edge AI Research and Engineering

Access the tools, guides, and training you need to build faster and smarter with MongoDB.

  • 0.1

    High accuracy

    Voyage AI models consistently rank atop the Retrieval Text Embedding Benchmark (RTEB)

  • 0.2

    Low dimensionality

    3x-8x shorter vectors ⇒ cheaper vector search and storage

  • 0.3

    Low latency

    4x smaller model and faster inference with superior accuracy

  • 0.4

    Cost efficient

    2x cheaper inference with superior accuracy

  • 0.5

    Long-context

    Longest commercial context length available (32K tokens)

  • 0.6

    Modularity

    Plug-and-play with any vectorDB and LLM

Deploy Anywhere

A resilient, cloud-first data layer

Everything you need to build modern applications, from database and search to analytics and AI, works together out of the box, so you ship features in days instead of stitching tools together for months.

“Harvey has partnered with Voyage AI to create custom legal embeddings tailored to Harvey’s use cases. We chose Voyage because their voyage-law-2 legal embedding model has already demonstrated best-in-class performance on public legal retrieval benchmarks. Fine-tuning the model with Harvey’s specific datasets reduces the irrelevant document rate by 25% and decreases the dimensionality of the embeddings and the vectorDB costs by threefold.”

Gabe Pereyra

President & Co-Founder

Gabe Pereyra, President & Co-Founder at Harvey