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 environments, the database provider has identified retrieval quality as a significant weakness that can negatively impact accuracy, cost-efficiency, and user confidence, even when the underlying AI models are performing optimally.

To address this issue, MongoDB launched four new versions of its embeddings and reranking models, collectively known as Voyage 4. These models are designed to enhance the efficiency and accuracy of data retrieval in AI applications. The Voyage 4 family includes voyage-4 embedding, a general-purpose model; voyage-4-large, considered MongoDB's flagship model; voyage-4-lite, optimized for low-latency and cost-sensitive applications; and voyage-4-nano, intended for local development, testing, and on-device data retrieval. Voyage-4-nano is also MongoDB's first open-weight model.

All Voyage 4 models are accessible through an API and on MongoDB's Atlas platform. According to MongoDB, these models outperform comparable models in the market.

The emphasis on retrieval quality highlights a growing concern within the AI community. While much attention is focused on the capabilities of large language models (LLMs), the effectiveness of these models is heavily dependent on the quality of the data they receive. RAG systems, for example, rely on retrieving relevant information from a database or knowledge base to augment the LLM's responses. If the retrieval process is flawed, the LLM may generate inaccurate or irrelevant outputs, undermining the entire system's reliability.

Agentic systems, which are designed to autonomously perform tasks, also depend on accurate data retrieval to make informed decisions. Poor retrieval quality can lead to errors, inefficiencies, and a lack of trust in the system's capabilities.

MongoDB's focus on embeddings and reranking models reflects a strategy to improve the precision and efficiency of data retrieval. Embeddings models transform data into numerical representations that capture semantic relationships, allowing for more accurate similarity searches. Reranking models further refine the search results by prioritizing the most relevant information.

The availability of different Voyage 4 models caters to a range of use cases, from general-purpose applications to specialized tasks requiring low latency or on-device processing. The release of an open-weight model, Voyage-4-nano, also signals a commitment to transparency and community collaboration.

The development underscores the importance of robust data infrastructure in the age of AI. As AI systems become more integrated into enterprise workflows, the need for reliable and efficient data retrieval will only continue to grow. MongoDB's latest offerings aim to address this critical need and contribute to the development of more trustworthy and effective AI applications.