Agentic systems and enterprise search depend on robust data retrieval that works efficiently and accurately. Database provider MongoDB touts its newest embedding model Moving more AI systems into production will help solve the declining retrieval quality.
As agentic and RAG systems move into production, retrieval quality is emerging as a silent failure point – which can undermine accuracy, cost, and user confidence even when the models themselves perform well.
The company launched four new versions of its embedding 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 stated that the Voyage-4 embedding serves as its general purpose model; MongoDB considers Voyage-4-Large as its flagship model. Voyage-4-Lite focuses on low-latency and low-cost tasks, and Voyage-4-Nano is for more local development and test environments or on-device data recovery.
Voyage-4-Nano is also MongoDB’s first open-weight model. All models are available via API and on MongoDB’s Atlas platform.
The company said the models outperform similar models from Google and Cohere on the RTEB benchmark. face hug rteb benchmark Keeps 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. “If you get them wrong, your search results will seem too random and superficial, but if you get them right, your application will suddenly feel like it understands your users and your data.”
He said the Voyage 4 model aims to improve the retrieval of real-world data, which often collapses after agentic and RAG pipelines go into production.
MongoDB also released a new multimodal embedding model, voyage-multimodal-3.5, which can handle documents including text, images, and video. This model vectorizes data and extracts semantic meaning from tables, graphics, figures, and slides commonly found in enterprise documents.
Enterprise embedding issues
For enterprises, an agent system is only as good as its ability to reliably get the right information at the right time. This requirement becomes harder as the workload scales and the context windows volume.
Many model providers target that layer of agentic AI. Google’s Gemini embedding model remained on top Embedding Leaderboard, and Cohere launches it 4 Embed Multimodal ModelWhich processes documents longer than 200 pages. Mistral said that its coding-embedding model, codestral embeddingOutperforms Cohere, Google and even MongoDB’s Voyage Code 3. MongoDB argues that benchmark performance alone does not address the operational complexity faced by enterprises in production.
MongoDB said that many customers have found that their data stacks cannot handle context-aware, retrieval-intensive workloads in production. The company said enterprises are seeing more fragmentation as they tie together disparate solutions to connect databases with recovery or reranking models. 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 no longer be treated as a loose collection of best-of-breed components. For enterprise agents to work reliably at scale, the embedding, reranking, and data layer need to work together as a tightly integrated system rather than a stitched-together stack.
<a href