At MongoDB.local San Francisco on January 15, 2026, MongoDB unveiled the integration of Voyage AI’s embedding and reranking models directly into its database platform, creating what the company calls a unified data intelligence layer. This expansion allows developers to query, embed and rerank data without exporting or duplicating it, cutting end-to-end latency by up to 40% on internal benchmarks and reducing operational complexity. The integration supports both on-premises MongoDB Community and cloud-based Atlas deployments via managed APIs and automated embedding pipelines.

MongoDB introduced five new embedding models from Voyage AI under the Voyage 4 series, designed to outperform public benchmarks such as the RTEB leaderboard. The flagship voyage-4-large model achieved a 3.7% improvement in retrieval F1 score over the nearest competitor in head-to-head tests, while the voyage-4-lite variant delivers 25% lower inference cost and 30% lower latency. An open-weight voyage-4-nano model supports local development and edge-device use cases. These additions position MongoDB as the first database vendor to offer production-grade embeddings optimized for cost, latency and accuracy in a single platform.

With the general availability of voyage-multimodal-3.5, MongoDB now supports unified embeddings for interleaved text, images and video—eliminating the need for custom parsers and multiple preprocessing steps. Automated Embedding for MongoDB Vector Search, in public preview, automatically generates high-fidelity vectors on data insert, update or query operations. Early tests at customers like TinyFish report a 50% reduction in engineering hours spent on pipeline maintenance, while maintaining vector freshness and retrieval accuracy even as datasets grow beyond 100 million documents.

More than 60,000 organizations run mission-critical workloads on MongoDB’s AI-ready platform, with Tavily and TinyFish cited as early adopters of the new Voyage 4 models. TinyFish CEO Sudheesh Nair noted that Python APIs for Voyage integration were ‘extremely lightweight and very fast,’ while Tavily’s Rotem Weiss praised the platform for letting his team focus on product innovation rather than infrastructure. MongoDB also launched an AI skills certification program to accelerate enterprise adoption, signaling its ambition to cement leadership in production AI data platforms.