Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What MongoDB states |
|---|---|
Deployment | Fully managed on Atlas, available on AWS, Google Cloud and Microsoft Azure in 125+ regions. Also generally available in self-managed Enterprise Advanced and Community Edition. |
Embeddings | Bring your own from any provider, or use Automated Embedding, which creates and syncs embeddings with Voyage AI models at ingest, update and query time. |
Search methods | Approximate (ANN, on HNSW graphs) and exact (ENN) nearest-neighbor search, and hybrid search that fuses lexical and vector results in one query. |
Self-managed license | MongoDB Search (mongot), the service that provides full-text and vector search, is free and source-available under the Server Side Public License (SSPL) v1. A Community Search Docker image is published. |
Our take
Choose Atlas Vector Search when your application data already lives in MongoDB and you want semantic or hybrid search on that live data without running a second database. If your data lives elsewhere, the no-sync benefit only arrives after you move operational data into MongoDB, so weigh it as a database decision, not just a vector index.
Evaluate MongoDB Atlas Vector Search
Popular comparisons
See all 7 MongoDB Atlas Vector Search comparisons
Where it stands out
One query across vectors and business data. Vector queries can be combined with metadata filters, graph lookups, aggregation pipelines, geospatial search and lexical search. Hybrid search weights can favor vector similarity or keyword relevance.
Less embedding plumbing. With Automated Embedding there is no external embedding service to run: MongoDB handles inference, authentication, scaling and retries, with the models hosted inside the Atlas security perimeter. Native reranking with Voyage AI reranker models runs inside the MongoDB query engine.
Separate scaling and smaller vectors. Dedicated Search Nodes scale vector search apart from the core database, which isolates the two workloads. Scalar and binary quantization, including automatic quantization of full-fidelity vectors, compresses vectors, which MongoDB says reduces storage and processing cost.
Product screenshot
Trade-offs to weigh
Exact search is for small sets. MongoDB says ENN can maintain sub-second latency for unfiltered queries up to 10,000 documents, and can give low-latency responses when highly selective filters narrow results to 10,000 documents or fewer. MongoDB positions approximate HNSW search for millions of vectors, and it trades some accuracy for performance.
A 4,096-dimension ceiling. Embeddings must stay under the service's 4,096-dimension limit, so check your embedding model's output size before you commit.
Self-managed enterprise use goes through sales. Enterprise Advanced, with its commercial license, is handled through MongoDB sales, and full-text and vector search for Enterprise Advanced clusters is an add-on.
Pricing and billing conditions
On Atlas, the main cost lines for vector search are the cluster that stores your data and indexes, any dedicated Search Nodes you add, and embedding model tokens if you use Automated Embedding, which MongoDB bills by actual token usage. With your own embeddings, your provider's billing, rate limits and API keys stay with you. Atlas clusters are pay-as-you-go, billed hourly with monthly invoices.
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Cluster option | Published price | What it includes |
|---|---|---|
Free (M0) | $0/hour, free forever | 512 MB storage, shared RAM and vCPU, up to 100 operations per second; for learning and exploring. Backups are not available. |
Flex | $0.011/hour, up to $30/month | 5 GB storage, shared RAM and vCPU; for development and testing, with some feature limitations. Billed by operations-per-second tier, from $0.0110/hour (0–100 ops/s, $8.00 for 30 days of continuous use) to $0.0411/hour (400–500 ops/s, $30.00). |
Dedicated | From $0.08/hour (M10); MongoDB states it starts at $56.94/month | 10 GB to 4 TB storage, 2 GB to 768 GB RAM, 2 to 96 vCPUs. MongoDB suggests M10 and M20 for pre-production and small non-critical workloads, and M30 and up for performance-critical workloads. |
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Tier | RAM and vCPUs | Listed hourly prices |
|---|---|---|
S20 | 4 GB RAM, 2 vCPUs | $0.12/hr, $0.14/hr or $0.15/hr |
S30 | 8 GB RAM, 4 vCPUs | $0.24/hr, $0.27/hr or $0.30/hr |
S40 | 16 GB RAM, 8 vCPUs | $0.48/hr, $0.57/hr or $0.62/hr |
S50 | 32 GB RAM, 16 vCPUs | $0.99/hr, $1.15/hr or $1.27/hr |
S60 | 64 GB RAM, 32 vCPUs | $1.77/hr, $2.22/hr or $2.29/hr |
S70 | 96 GB RAM, 48 vCPUs | $2.50/hr or $3.23/hr |
S80 | 128 GB RAM, 64 vCPUs | $3.26/hr or $4.22/hr |
MongoDB lists more than one hourly price, with a different storage size per node, for most Search Node tiers, and Atlas costs vary by cloud provider and region. Confirm the price for your provider and region before sizing Search Nodes.
Other cost drivers: storage counts both data and indexes; data leaving the cluster or moving between regions or providers is charged, though Free and Flex tiers are not charged for outgoing data; backups bill by region and snapshot storage.
Credits: MongoDB for Startups offers free Atlas credits, educators can get $500 in Atlas credits, and students $50 through the GitHub Student Developer Pack.
Public signals
About these signals
Verified factual signals from public sources. They indicate observable activity or interest, not total adoption, product quality, or cost.
609 GitHub commits 90d159 GitHub stars
See all signals from 4 sources
Three-month score against stable baseline terms—not search volume or adoption.
Alternatives to MongoDB Atlas Vector Search
The reviewed substitutes for MongoDB Atlas Vector Search among the vector databases, and what would make each one the better answer.
Direct alternatives
Reviewed substitutes: products bought for the same job, where a team picks one.
- Milvus
- Milvus and MongoDB Atlas Vector Search give RAG and search applications a place to index embeddings and query them by similarity. Milvus is a dedicated open-source vector database that can be self-hosted or bought as a managed service, while MongoDB Atlas Vector Search is built into the managed MongoDB Atlas service so vectors stay next to the application's documents.Applies to: Choosing between these two for the vector databases decision.
- Redis Vector Search
- Teams already on Redis or MongoDB can keep embeddings in that database and query them by similarity with Redis Vector Search or MongoDB Atlas Vector Search. Redis Vector Search answers hybrid queries from memory for very low latency, while MongoDB Atlas Vector Search runs in managed MongoDB Atlas and stores vectors in the same documents as the rest of the data.Applies to: Choosing between these two for the vector databases decision.
- Weaviate
- Weaviate and MongoDB Atlas Vector Search both keep embeddings next to the data they describe and answer semantic search queries for AI applications. Weaviate is a dedicated open-source vector database that can be self-hosted or used as a managed cloud, while MongoDB Atlas Vector Search is a feature of the managed MongoDB Atlas document database.Applies to: Choosing between these two for the vector databases decision.
- Qdrant
- Qdrant and MongoDB Atlas Vector Search both index embeddings and answer similarity queries for RAG applications. Qdrant is a dedicated open-source vector search engine written in Rust, while MongoDB Atlas Vector Search is part of the managed MongoDB Atlas database, so embeddings are stored in the same documents as operational data.Applies to: Choosing between these two for the vector databases decision.
- Aerospike
- MongoDB Atlas Vector Search and Aerospike add vector search to an operational database, so embeddings live in the same system as application data. MongoDB Atlas Vector Search is part of the managed MongoDB document service, while Aerospike is a multi-model key-value and document database built for real-time workloads with predictable low latency at large scale.Applies to: Choosing a vector store for embedding search in a retrieval or agent application.
Other approaches
A different approach to the same problem. Each substitutes only for the workload named beside it.
- Marqo
- Marqo optimises search conversion using click-stream, purchase and event data, creating a personalised experience that knows what your customers are looking for - better than they do.Applies to: For teams that build their own product search on a vector store rather than buying a hosted ecommerce search platform.
- Pinecone
- Pinecone and MongoDB Atlas Vector Search both run similarity search over embeddings for AI features. Atlas Vector Search adds vector indexes to documents a team already keeps in MongoDB Atlas, while Pinecone is a dedicated managed vector database, chosen when vector search is a workload of its own.Applies to: Whether embedding search can reuse an existing datastore or needs a dedicated vector service.
- ChromaDB
- ChromaDB is chosen instead of MongoDB Atlas Vector Search for Python-centric RAG prototypes and embedding-first retrieval workloads. We recommend ChromaDB for data and AI teams building early RAG applications with LangChain or LlamaIndex that want a lightweight starting point.