Voyage AI (MongoDB): Foundation Models

Voyage AI builds state-of-the-art embedding models and rerankers that power search, retrieval, and RAG applications. Founded in September 2023 by Stanford CS Assistant Professor Tengyu Ma, with advisors including Fei-Fei Li, Christopher Manning, and Christopher Re, Voyage AI quickly established itself as a leader in the embedding space by consistently outperforming OpenAI and Cohere on retrieval benchmarks.

The company raised $28M in total funding including a Series A led by CRV, with participation from Snowflake and Databricks. In February 2025, MongoDB acquired Voyage AI for $220M in cash and stock — a remarkable exit after less than two years of operation and a 7.8x return on invested capital. The product and API continue to operate under the Voyage AI brand as part of MongoDB's data platform.

Voyage AI's model lineup includes the Voyage 4 family (launched January 2026), which introduced the first production-grade Mixture of Experts (MoE) architecture for embeddings. The family features shared embedding spaces across model sizes, enabling teams to use cheaper models for queries and more expensive ones for indexing. Domain-specific models for code, law, and finance outperform general-purpose alternatives by ~15% on domain tasks, and the 32K context window is 4x OpenAI's limit.