MongoDB integrates Voyage 4 models across its platform

MongoDB said it is integrating its Voyage 4 embedding and reranking models into its platform infrastructure in a move that will improve accuracy, deliver better context and help developers scale applications.

With the moves, announced at MongoDB.local San Francisco, MongoDB is looking to grab a bigger piece of the AI applications pie. MongoDB is also creating a unified data intelligence layer for production AI in a move that will enhance its enterprise footprint.

MongoDB's Ben Cefalo, SVP, Head of Core Products and Atlas Foundational Services, said the company has worked with customers to understand where things break as AI moves from prototype to production. "Those conversations rarely start with AI models. They start with really practical questions like, how do you get our data ready? How do we keep things performant as we scale? How do we ensure accuracy of results? How do we avoid gluing together five different systems," said Cefalo. "What's interesting is that customers increasingly do not think of MongoDB as just a database. They reframe the database as a foundation for their AI stack. They are increasingly thinking of MongoDB as their strategic data platform."

• MongoDB Q3 surges on Atlas demand
• MongoDB names CJ Desai CEO
• MongoDB brings full-text search, vector search to on-prem, self-managed versions
• MongoDB launches effort to modernize legacy data, application layers

Cefalo added that the main goals of developers are to scale and operate applications at scale with minimized risk of hallucinations without the need to move data. MongoDB is unifying databases, vector stories and model APIs on one platform.

Here's what MongoDB announced:

• New embedding models from Voyage AI, which was acquired a year ago and serves as MongoDB's embedding and retrieval model suite. MongoDB said its Voyage 4 models are generally available including the general purpose voyage-4 embedding model, voyage-4-large for retrieval accuracy, voyage-4-lite for optimized latency and cost and the open weight voyage-4-nano for on-device applications.
• Voyage-multimodal-3.5 is generally available to extract context from video, images and text. The model vectorizes multimodal data to capture semantic meaning from various documents.
• MongoDB's Atlas Embedding and Reranking API expose Voyage AI models natively within Atlas.
• Auto Embeddings for MongoDB Vector Search, which automatically generates and stores embeddings using Voyage AI whenever data is inserted, queried or changed. The feature eliminates the need for separate embedding pipelines and external model services. Auto embedding is in public preview for MongoDB drivers in various frameworks, available in MongoDB Community and coming to MongoDB Atlas soon.
• An intelligent assistant for MongoDB Compass and Atlas Data Explorer is now generally available.

Separately, MongoDB expanded its MongoDB for Startups program, which is designed to enable startups to scale on its platform. Initial launch partners include Temporal and Fireworks AI.