AI MODELS
Build trustworthy, production-grade AI applications and agents with best-in-class embedding models—accessible via standalone API or natively in Atlas.
Maximize your retrieval performance with state-of-the-art accuracy that reduces hallucinations across complex and multimodal data.
Ground your AI applications and agents with high-accuracy retrieval, powered by state-of-the-art models and advanced capabilities like contextualized chunk embeddings and instruction following.
Refine your results to maximize retrieval quality. Leverage the power of cross-encoders capable of instruction following.
Project text, screenshots, PDFs, and video into a single unified vector space. Designed for systems architects and engineering teams building workflows that must look beyond raw text to parse complex, multimodal documents natively without loss of semantic context.
Start with the flexibility of a database-agnostic API that fits your stack today. As your workflows mature and your needs grow, seamlessly adopt integrated capabilities that simplify operations and reduce coordination across services—secure and governed from day one.
Cutting-edge research delivers models at the price-performance frontier. Right-size inference and storage costs without compromising retrieval quality through flexible output dimensions, quantization, and a range of model sizes tailored to your requirements.
Overcome infrastructure constraints with flexible deployment controls designed for multi-geo and privacy-first operations. Run workloads seamlessly on standard managed cloud APIs, private VPCs, or air-gapped on-premises infrastructure to meet strict compliance mandates.




No. Voyage AI is built for agnostic utility. While we offer native, high-performance integration for MongoDB Atlas users, our models are available via a standalone API. You can integrate Voyage AI’s high-precision embeddings and rerankers with any database or existing AI framework without vendor lock-in.
Voyage AI utilizes a dedicated third embedding queue engineered explicitly for large-scale auto-embedding. This ensures that large initial sync jobs (up to billions of documents) finish rapidly without affecting the latency or availability of your live application requests.