Voyage AIUnclaimed

Voyage AI provides embedding models and rerankers that help applications search and retrieve information from unstructured data. The page describes them as the retrieval layer behind retrieval-augmented generation (RAG), where relevant documents are found and passed to a language model to produce factual answers.

The models are offered as an API and are available through the major cloud providers and data platforms. The page also lists SaaS and customer tenant deployment inside a customer's VPC, and custom or on-premise deployments with model licensing on request.

According to the page, the models are designed for high accuracy, with shorter vectors that make vector search and storage cheaper, and with a context length of 32K tokens. It says they work with any vector database and any LLM. The company reports that its models rank at the top of the Retrieval Text Embedding Benchmark (RTEB), which it describes as built to reflect enterprise retrieval rather than academic datasets.

The page is aimed at builders and enterprises that need search over documents, legal or other domain text, and AI applications that depend on retrieval quality. One customer quoted on the page, Harvey, says it uses a fine-tuned version of the legal embedding model for its own datasets.

Pricing is consumption-based, with pay-as-you-go access through the cloud providers. Custom and on-premise deployments are priced by contact. Documentation, a quick start tutorial and guides are linked from the site.

Voyage AI is part of MongoDB, and the page notes that the models are available in MongoDB's Intelligent Data Platform.