MongoDB, Inc. announced an expansion of its AI capabilities at MongoDB. local San Francisco, bringing together its core database with Voyage AI's world-class embedding and reranking models to deliver a unified data intelligence layer for production AI. By integrating these models directly into MongoDB's platform infrastructure, developers can now build and operate sophisticated applications at scale with reduced risk of hallucinations, without the need to move or duplicate data.
To support developers moving AI applications into production, MongoDB introduced a set of new AI capabilities designed to simplify how intelligent applications are built and operated. The company unveiled five embedding models from Voyage AI, MongoDB's embedding and retrieval model suite, Automated Embedding for MongoDB Community Vector Search, embedding and reranking AI model APIs in Atlas, and an AI-powered data operations assistant for MongoDB Compass and Atlas Data Explorer. These capabilities strengthen MongoDB's position as the leading AI-ready data platform, trusted by more than 60,000 customers running mission-critical workloads.
Voyage AI models are available through MongoDB Atlas via API, integrated with MongoDB Community through managed Automated Embedding, and remain fully available as a standalone platform independent of MongoDB. Instead of stitching together an operational database, a vector store, and multiple pipelines, teams can keep operational data and retrieval capabilities together, reducing latency and synchronization overhead. The result is a simpler architecture, faster iteration, and AI applications that are built to run reliably in production, not just in demos.
New capabilities include: accuracy with models from Voyage AI: The general availability of the new Voyage 4 series continues giving developers high performing embedding models--which outperform Gemini and Cohere on the public RTEB leaderboard--for more accurate retrieval at lower cost. The Voyage 4 series includes the general-purpose voyage-4 embedding model, which strikes a balance between retrieval accuracy, cost, and latency, the flagship voyage-4-large model for the highest retrieval accuracy, voyage-4-lite for optimized latency and cost, and an open-weights voyage-4-nano for local development and testing, or on-device applications. Facilitated context extraction from video, images, and text: The general availability of the New Voyage-multimodal-3.5 model expands support for interleaved text and images to now include video.
Voyage AI's voyage-multimodal- 3 was the first production-grade embedding model to handle interleaved text and images, voyage-multimodal -3.5 advances this unified processing approach, more effectively vectorizing multimodal data together to best capture key semantic meaning from tables, graphics, figures, slides, PDFs, and more. By handling embedding generation natively within the database, MongoDB removes the need for separate embedding pipelines or external model services. Embeddings stay fresh as data changes, helping retrieval to remain accurate and AI applications to maintain reliable context.
The result is a simpler Architecture with fewer moving parts, making it easier for teams to build and run AI-enabled applications in production. Automated Embedding is available in public preview with support in drivers (e.g. Javascript, Python, Java, etc) and AI Frameworks like LangChain and LangGraph (Python). Available for MongoDB Community, and coming soon on MongoDB Atlas.
MongoDB's Atlas Embedding and Reranking API exposes Voyage AI models natively within Atlas, allowing teams to ship AI features with enterprise-grade security, performance, and reliability infrastructure. An intelligent assistant for MongoDB Compass and At Atlas Data Explorer is now generally available, delivering natural-language, AI-powered assistance for everyday data operations, such as query optimization. MongoDB also introduced a new AI skills certification to help teams scale scale and workloads.
