Voyage AI
Voyage AI develops embedding and reranking models for search. Founded in 2023, it became part of MongoDB in February 2025. Its catalog includes hosted retrieval models and the downloadable Voyage 4 Nano checkpoint.
Company & leadership
- Acquired entity
- Voyage AI Innovations, Inc.
- Owner
- MongoDB acquired all outstanding shares on 17 February 2025; announced on 24 February
- Parent headquarters
- New York, United States
- Research leadership, January 2026
- Tengyu Ma was listed as Chief AI Scientist at MongoDB
- Parent company leadership
- Dev Ittycheria became interim President and CEO on 28 September 2026
Documented co-founders
- Tengyu Ma, Kaidi Cao and Hong Liu
Origins and research
Voyage began in 2023 with a focus on the retrieval step in AI applications. Its first public embedding announcement introduced voyage-01. The team drew on research in contrastive learning, which trains representations by comparing related and unrelated examples.
The founding record spans several primary sources. MongoDB identifies Tengyu Ma as Voyage's founder. Kaidi Cao's Stanford profile says that he co-founded Voyage and helped lead its acquisition. Tengyu Ma's research-group page identifies Hong Liu as a Voyage co-founder.
Funding and MongoDB ownership
In October 2024, Voyage announced a US$20 million Series A led by CRV. The announcement put total funding at US$28 million. These amounts describe investment raised, not a company valuation.
MongoDB's quarterly filing records the acquisition on 17 February 2025. The public announcement followed a week later. It described bringing Voyage's retrieval models into MongoDB while retaining access through Voyage and cloud services.
Leadership titles here have a defined scope and date. The January 2026 MongoDB event named Tengyu Ma as Chief AI Scientist. In September 2026, MongoDB named Dev Ittycheria interim President and CEO after CJ Desai left. Neither source establishes a separate current Voyage CEO.
Catalog and access
MongoDB's lifecycle policy covers its Atlas model service. It does not cover api.voyageai.com, self-managed models or marketplace services. Atlas Legacy models remain served; that label does not mean retirement.
Voyage also lists earlier API models. Deprecated entries retain that status below. Tokenizer repositories do not provide model weights.
Nano's official GGUF conversions use the same trained model. The experimental voyage-3-m-exp release uses benchmark-oriented training.