MongoDB believes that improved data retrieval, rather than simply larger AI models, is crucial for building trustworthy enterprise AI systems. As agentic systems and Retrieval-Augmented Generation (RAG) move into production, the database provider has identified retrieval quality as a significant weakness that can negatively impact accuracy, cost, and user trust, even if the underlying models are strong.

To address this, MongoDB launched four new versions of its embeddings and reranking models, all under the "Voyage 4" umbrella. These models are designed to improve the efficiency and accuracy of data retrieval in AI systems. Voyage 4 is available in four modes: voyage-4 embedding, voyage-4-large, voyage-4-lite, and voyage-4-nano.

According to MongoDB, voyage-4 embedding serves as its general-purpose model, while Voyage-4-large is considered its flagship model. Voyage-4-lite is optimized for tasks requiring low latency and reduced costs. Voyage-4-nano is intended for local development and testing environments, as well as on-device data retrieval. Notably, voyage-4-nano is MongoDB's first open-weight model.

All four models are accessible through an API and on MongoDB's Atlas platform. The company claims that these models outperform similar models currently available.

The focus on retrieval quality highlights a growing concern in the AI industry. While large language models (LLMs) have garnered significant attention, the ability to effectively retrieve relevant information from databases and knowledge bases is essential for building reliable and accurate AI applications. RAG systems, in particular, rely on accurate retrieval to augment the knowledge of LLMs with external data.

The implications of poor retrieval quality extend beyond mere inaccuracies. Inaccurate or inefficient retrieval can lead to increased costs due to wasted computational resources and can erode user trust in AI systems. As AI becomes more integrated into critical business processes, ensuring the reliability of data retrieval is paramount.

MongoDB's emphasis on retrieval quality suggests a shift in focus within the AI community. Rather than solely pursuing larger and more complex models, companies are beginning to recognize the importance of optimizing the entire AI pipeline, including data retrieval. This holistic approach is essential for building AI systems that are not only powerful but also trustworthy and cost-effective.

The availability of MongoDB's Voyage 4 models represents a step toward addressing the challenges of retrieval quality in enterprise AI. The company plans to continue developing and refining its models to meet the evolving needs of the AI industry.