MongoDB Atlas Vector Search

MongoDB Atlas Vector Search is a native vector database capability built directly into MongoDB Atlas, letting developers store and query vector embeddings alongside their existing operational data rather than standing up and syncing a separate, standalone vector database. MongoDB was founded in 2007 in New York City as 10gen by Dwight Merriman, Eliot Horowitz, and Kevin Ryan, three veterans of the ad-tech company DoubleClick who had personally hit the scaling limits of relational databases while serving over 400,000 ads per second. The company renamed itself MongoDB in 2013, launched its fully managed cloud platform, Atlas, in 2016, and went public on Nasdaq under the ticker MDB in 2017, reporting roughly $2.46 billion in annual revenue and over 5,600 employees by 2026.

Atlas Vector Search's core pitch is architectural simplicity: because vector embeddings live in the same document as the rest of an application's data, teams avoid the "synchronization tax" of keeping a separate vector database in sync with a primary datastore. The platform supports both Approximate Nearest Neighbor (ANN) search using a Hierarchical Navigable Small World (HNSW) graph for large-scale performance, and Exact Nearest Neighbor (ENN) search for cases requiring perfect accuracy on smaller result sets. Hybrid search lets developers combine lexical (keyword) and vector search in a single query, and native reranking, powered by Voyage AI (which MongoDB acquired), improves retrieval precision directly within the query engine, reducing token costs and latency for LLM-based applications.

More recent additions extend the platform further into applied AI workflows: Automated Embedding, powered by Voyage AI, generates and syncs vector embeddings directly in the database with no separate ML pipeline required, and dedicated Search Nodes let teams scale vector search compute independently from their core transactional workload, an architectural advantage MongoDB positions as superior to competitors that can't isolate search performance from database performance. The platform supports embeddings from any provider under a 4096-dimension limit, along with scalar and binary vector quantization for more efficient storage. Customers using Atlas Vector Search in production include Novo Nordisk (cutting clinical report creation time to 10 minutes), Okta (30% lower operating costs), and Delivery Hero for real-time product recommendations.

Pricing follows Atlas's broader consumption-based model: Atlas Search and Vector Search run at no additional cost on shared database nodes, while production workloads needing dedicated performance can add Search Nodes (minimum two, requiring an M10+ cluster) starting around $0.12/hour. A free M0 tier with 512MB of storage supports basic vector search for prototyping RAG applications and semantic search at zero cost, making Atlas Vector Search well suited for teams already using or willing to adopt MongoDB as their primary database who want to avoid managing a second, standalone vector search system, though teams specifically seeking the most mature, dedicated vector-only infrastructure at very large scale should still compare it against purpose-built vector databases.

MongoDB Atlas Vector Search is not a separately priced product; it runs on your existing Atlas cluster resources. The Free tier (M0) offers 512MB of storage and supports basic vector search for learning and prototyping at no cost. The Flex tier costs between $8 and $30/month (usage-capped, pay-as-you-go hourly), suitable for low-to-moderate production workloads and replacing the deprecated Shared and Serverless tiers as of February 2025. Dedicated clusters start at $57/month for entry-level production workloads, with hourly rates that vary by cloud provider (AWS, Azure, GCP) and region. For production workloads needing isolated vector search performance, dedicated Search Nodes (requiring a minimum of two nodes on an M10+ cluster) are billed separately, ranging from roughly $0.12/hour for an S20 node up to $4.22/hour for the largest S80 tier. Data transfer/egress is billed separately at standard per-GB cloud provider rates, and self-managed Enterprise Advanced deployments require a custom sales quote. Serverless instances were fully retired on January 22, 2026, with existing customers migrated to Free, Flex, or Dedicated tiers.