MongoDB launches Voyage 4 embedding models to fix enterprise AI retrieval problems at scale

Why MongoDB thinks better retrieval -- not bigger models -- is the key to trustworthy enterprise AI

Agentic systems and enterprise search depend on strong data retrieval that works efficiently and accurately. Database provider MongoDB thinks its newest embeddings models help solve falling retrieval quality as more AI systems go into production. As agentic and RAG systems move into production, retrieval quality is emerging as a quiet failure point -- one that can undermine accuracy, cost, and user trust even when models themselves perform well. The company launched four new versions of its embeddings and reranking models. Voyage 4 will be available in four modes: voyage-4 embedding, voyage-4-large, voyage-4-lite, and voyage-4-nano. MongoDB said the voyage-4 embedding serves as its general-purpose model; MongoDB considers Voyage-4-large its flagship model. Voyage-4-lite focuses on tasks requiring little latency and lower costs, and voyage-4-nano is intended for more local development and testing environments or for on-device data retrieval. Voyage-4-nano is also MongoDB's first open-weight model. All models are available via an API and on MongoDB's Atlas platform. The company said the models outperform similar models from Google and Cohere on the RTEB benchmark. Hugging Face's RTEB benchmark puts Voyage 4 as the top embedding model. "Embedding models are one of those invisible choices that can really make or break AI experiences," Frank Liu, product manager at MongoDB, said in a briefing. "You get them wrong, your search results will feel pretty random and shallow, but if you get them right, your application suddenly feels like it understands your users and your data." He added that the goal of the Voyage 4 models is to improve the retrieval of real-world data, which often collapses once agentic and RAG pipelines go into production. MongoDB also released a new multimodal embedding model, voyage-multimodal-3.5, that can handle documents that include text, images, and video. This model vectorizes the data and extracts semantic meaning from the tables, graphics, figures, and slides typically found in enterprise documents. Enterprise's embeddings problems For enterprises, an agentic system is only as good as its ability to reliably retrieve the right information at the right time. This requirement becomes harder as workloads scale and context windows fragment. Several model providers target that layer of agentic AI. Google's Gemini Embedding model topped the embedding leaderboards, and Cohere launched its Embed 4 multimodal model, which processes documents more than 200 pages long. Mistral said its coding-embedding model, Codestral Embedding, outperforms Cohere, Google, and even MongoDB's Voyage Code 3. MongoDB argues that benchmark performance alone doesn't address the operational complexity enterprises face in production. MongoDB said many clients have found that their data stacks cannot handle context-aware, retrieval-intensive workloads in production. The company said it's seeing more fragmentation with enterprises having to stitch together different solutions to connect databases with a retrieval or reranking model. To help customers who don't want fragmented solutions, the company is offering its models through a single data platform, Atlas. MongoDB's bet is that retrieval can't be treated as a loose collection of best-of-breed components anymore. For enterprise agents to work reliably at scale, embeddings, reranking, and the data layer need to operate as a tightly integrated system rather than a stitched-together stack.

MongoDB combines database and embedding models for simplified AI development - SiliconANGLE

MongoDB combines database and embedding models for simplified AI development MongoDB Inc. is making its play for the hearts and minds of artificial intelligence developers and entrepreneurs with today's announcement of a series of new capabilities designed to help developers move applications from prototype to production more quickly. They include the general availability of the Voyage 4 family of embedding models and a planned expansion of the MongoDB for Startups program. The new features tighten the integration between MongoDB's core database platform and the retrieval and embedding technologies it acquired with the purchase of Voyage AI Inc. last year. Embeddings are numerical representations of data that capture semantic meaning as vectors. They allow systems to compare and retrieve information based on meaning rather than exact keywords, which is essential for many AI tasks. "Customers increasingly do not think of MongoDB as just a database; they reframe the database as a foundation for their AI stack," said Benjamin Flast, director of product management. The Voyage 4 series of embedding models are now available through application programming interfaces in the MongoDB Atlas managed service and can also be used in the on-premises community edition of MongoDB. The lineup includes multiple models designed to balance retrieval accuracy, latency and cost. They include voyage-4 for general-purpose use, voyage-4-large for maximum retrieval accuracy, voyage-4-lite for lower latency and cost and voyage-4-nano, an open-weights model intended for local development and testing. MongoDB said the models are designed to improve retrieval accuracy for production artificial intelligence workloads by reducing the need to move or duplicate data across separate systems. The company also announced the general availability of voyage-multimodal-3.5, which expands support for interleaved text and images to include video. The model is intended to simplify context extraction from complex documents and multimedia sources. "This unlocks unified retrieval across multiple content types," said Franklin Sun, staff product manager at MongoDB. "You have one embedding model instead of three to handle different data types. You also have better end-user experiences where the system can understand the relationship between what someone wrote, what they saw and what they recorded." MongoDB said tighter integration between its operational database and retrieval models allows developers to avoid managing separate vector databases, pipelines and synchronization processes, which can introduce latency and operational risk. The company also introduced automated embedding capabilities for MongoDB Community Vector Search, now available in public preview. The feature automatically generates and stores embeddings whenever data is inserted, updated or queried, eliminating the need for separate embedding pipelines or external services. Automated embedding is available today for MongoDB Community Edition and is expected to be available soon on the Atlas service. MongoDB said the feature integrates with its drivers and artificial intelligence frameworks such as LangChain and LangGraph. For Atlas users, MongoDB also introduced embedding and reranking APIs that expose Voyage AI models directly within the platform. In addition, an artificial intelligence-powered assistant for MongoDB Compass and Atlas Data Explorer is now generally available. The assistant provides natural-language guidance for common data operations such as query optimization and troubleshooting. MongoDB for Startups, a program that helps early-stage companies scale applications from initial development to global deployment, is getting a boost with an expanded partner ecosystem. The company said startups participating in the program now represent more than $200 billion in combined valuation, based on Pitchbook data from last month. Through the program, eligible startups can access matched credits, coordinated onboarding and joint enablement resources across participating technologies. Initial launch partners include Fireworks AI Inc. and Temporal Technologies Inc. Samar Abbas, chief executive officer of Temporal, said the partnership is intended to simplify distributed application development. "This allows us to reach a community of developers who value a strong data foundation," he said in a statement. MongoDB said additional partners and offerings are expected to be added to the startup program over time.

MongoDB Aims For Production-Ready AI Apps With New Model Capabilities

MongoDB is more tightly integrating the embedding and reranking model technology it obtained last year through its Voyage AI acquisition with its database development platform. MongoDB is expanding the AI capabilities of its database and application development platform with newly integrated embedding and reranking models that the company says will improve the accuracy of AI applications as they move from development into production. The new functionality, based on technology MongoDB acquired when it bought Voyage AI in February 2025, delivers a unified data intelligence layer for production AI and helps developers build and operate sophisticated AI applications at scale with minimized risk of hallucinations and without the need to move or duplicate data, according to the company. MongoDB, which is holding its MongoDB.local San Francisco event today, also announced an expansion of MongoDB for Startups, a program that provides startup companies with technical expertise, financial credits and other resources to build AI software using a technology stack with MongoDB Atlas as its foundation. [Related: MongoDB Names Former Cloudflare Exec To Take Over As CEO] "Over the past few months, we've spent time with countless customers, founders, executives at large companies, platform teams, developers -- not to pitch but to understand where things break as AI moves from prototype to production," said Ben Cefalo, senior vice president, MongoDB head of core products and Atlas Foundational Services, in a press briefing prior to today's MongoDB.local San Francisco event. "Those conversations rarely start with AI models. They start with really practical questions like 'How do we 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 or extensions just to ship something? What's going to be the ROI?'" Cefalo said. While MongoDB is generally seen as among the leading next-generation database systems, the company in recent years has positioned its software -- especially the MongoDB Atlas cloud-native database -- as a foundation for the technology stack needed to develop and run AI applications. In 2024 the company launched the MongoDB AI Applications Program (MAAP) through which the company partners with the cloud hyperscalers, large language model developers, AI development tool providers, and system integrators and consulting partners to provide a technology stack and reference architectures for building AI systems. Cefalo pointed to capabilities within the MongoDB platform that support production AI systems, at scale, including support for structured, semi-structured and unstructured data; real-time operational data; data accuracy controls; vector search and hybrid search functionality; and automated embedding. "At the end of the day, all this work comes back to one thing: Helping builders build," Cefalo said. "The database, the platform, the industry-leading AI capabilities -- it's all in the service of turning ideas into systems that actually run." Integrating Voyage AI's embedding and reranking models with the MongoDB core database provides a unified data intelligence layer for production AI, according to the company. With the addition of these models into the MongoDB platform infrastructure developers can build and operate complex AI applications at scale, with reduced risk of hallucinations, and without the need to move or duplicate data. MongoDB announced the general availability of a new Voyage 4 model series including the general-purpose Voyage-4 embedding model, the flagship Voyage-4 large language model for retrieval accuracy, Voyage-4-lite, and the open-weights Voyage-4-nano for local development and testing and for on-device applications. Also now generally available is the new Voyage-Multimodal-3.5 model with expanded support for interleaved text and images. The company also unveiled a public preview of Automated Embedding for MongoDB Vector Search in the MongoDB Community edition with availability soon in MongoDB Atlas. Also now generally available is the Atlas Embedding and Reranking API that exposes Voyage AI models natively within Atlas. "This announcement is all about taking one of the most painful parts of building AI applications, that is, managing retrieval and managing embedding models, and pushing it down into the database," said Frank Liu, a Voyage AI product manager, during the press conference. "For builders, this means that there is one place to manage all of your data, your embeddings and your retrieval. There's less friction from getting [an] idea to a working AI feature and also a clean path from prototype to production using the same models and platform," Liu said. "All this is designed to pair you with the rest of MongoDB's capabilities like auto embedding, vector search, and a more unified developer experience, so that when you choose MongoDB for your AI workloads, you're not just getting a database, you're getting a retrieval stack that can keep up with your ambitions as a developer." MongoDB said the companies that participate in its MongoDB for Startups program now represent more than $200 billion in combined valuations. The overall goal of the program is to provide startup companies with a complete infrastructure stack, so startups can avoid having to devote time to infrastructure decisions. Under the expansion the program's ecosystem of supporting IT vendors, including AI workflow platform developer Temporal and generative AI platform provider Fireworks AI, will provide startups with match credits, enabling content, joint events and other benefits across complementary technologies. "The goal is to turn MongoDB for Startups from a one-way perks marketplace to a two-way ecosystem, where partners and startups both benefit from being in the program," said Suraj Patel, vice president of MongoDB Ventures & Corporate Development, during the press conference. "This reciprocal partnership with MongoDB allows us to reach a community of developers who value a strong data foundation," said Temporal CEO Samar Abbas, in a statement. "We look forward to creating a collaborative ecosystem that simplifies complexity for founders as they push the boundaries of distributed systems and workflow orchestration." "By joining this program, we are ensuring [startup] founders who choose MongoDB can easily access our high-performance inference engine, creating a seamless path to scale their AI ambitions together," said Lin Qiao, Fireworks co-founder and CEO, also in a statement. MongoDB is also putting increased emphasis on recruiting more startups from San Francisco and the Bay Area, with plans to host more than 50 local events in the next year, with a particular emphasis on AI startups.