At a glance
| Product type | Standalone open source vector database server, plus managed Qdrant Cloud |
| Vector index types | HNSW (filterable); sparse vectors; quantization including TurboQuant |
| Hybrid search | Yes: Query API prefetch with RRF or DBSF fusion; server-side BM25 |
| Metadata filtering | Yes: JSON payload filters with typed payload indexes (keyword, numeric, geo, datetime, text, uuid) |
| Embedding integration | FastEmbed (client library); Cloud Inference on Qdrant Cloud, BM25 free on all tiers |
| Scaling and multitenancy | Sharding and replication; tenant partitioning, user-defined shards or tiered multitenancy |
| Deployment (managed or self-hosted) | Self-hosted (Docker, Kubernetes), Qdrant Cloud, Hybrid Cloud, Private Cloud |
| Licensing and pricing model | Apache 2.0; cloud free tier, then hourly usage on vCPU, RAM and storage |
Qdrant is a strong pick when search results must respect rich filters, because its HNSW index can be made filter-aware and its payload indexes cover keyword, numeric, geo, datetime and text fields. It is free to self-host; on Qdrant Cloud the free cluster is for prototypes only, and Standard pricing is resource-based with rates shown only in the calculator.
Overview
Qdrant is a standalone vector database server written in Rust and released under the Apache 2.0 licence. Each point holds one or more vectors (dense, sparse or multi-vector) and a JSON payload. You can run it yourself with Docker or on Kubernetes, use the managed Qdrant Cloud on AWS, Google Cloud or Azure, or run Qdrant-managed clusters on your own infrastructure with Hybrid Cloud. Recent releases include 1.17 (relevance feedback, weighted RRF, audit logging), 1.18 (TurboQuant quantization) and 1.19 (5 August 2026: memory tiers, a 4-bit TurboQuant storage datatype, per-tenant IDF statistics for BM25, and removal of the legacy /search, /recommend and /discover endpoints in favour of /query).
The vector index is HNSW. Qdrant's distinguishing feature is filterable HNSW: if payload indexes exist before the graph is built, Qdrant adds filter-aware edges so filtered searches stay accurate. Hybrid search runs through the Query API, which prefetches dense and sparse (for example BM25) results and fuses them with RRF or DBSF; since 1.15.2 Qdrant converts text to BM25 sparse vectors on the server. Dense embeddings can be produced client-side with FastEmbed or, on Qdrant Cloud, by Cloud Inference.
Key features
- HNSW vector index with filter-aware graph edges when payload indexes are created before indexing
- Payload indexes for keyword, integer, float, bool, geo, datetime, text and uuid fields, plus a full-text index with optional BM25 ranking
- Query API for hybrid search: prefetch dense and sparse results and fuse them with RRF, weighted RRF or DBSF
- Server-side BM25 sparse vector conversion (since 1.15.2) and Cloud Inference for hosted embedding models on Qdrant Cloud
- Multitenancy by payload partitioning (is_tenant), user-defined sharding or tiered multitenancy that promotes large tenants to dedicated shards
- Quantization options for memory reduction, including TurboQuant added in 1.18
- Deployment choice: self-hosted, Qdrant Cloud, Hybrid Cloud on your own infrastructure, or Private Cloud
Integrations and ecosystem
Pros and cons
Pros
- Apache 2.0 licence, written in Rust, with documented Docker and Kubernetes deployment
- Filter-aware HNSW and typed payload indexes for accurate filtered vector search
- Hybrid search with server-side BM25 and documented fusion methods
- Several documented multitenancy patterns, from shared collections to dedicated shards per large tenant
Cons
- Payload indexes must exist before HNSW is built to get filter-aware edges; adding them later means rebuilding the index
- Free cloud cluster has no high availability and is deleted after 4 weeks of inactivity
- Standard tier unit rates are not printed on the pricing page, only in the calculator
- Version 1.19 removed the legacy /search, /recommend and /discover endpoints, so older clients must migrate to /query
Pricing
The open source database is free to self-host. Qdrant Cloud has a free-forever tier: a single-node cluster with 0.5 vCPU, 1 GB RAM and 4 GB disk (about 1 million 768-dimension vectors), no high availability, suspended after 1 week unused and deleted after 4 weeks of inactivity if not reactivated. The Standard tier is usage-based, billed hourly on vCPU, memory and storage for clusters, backup storage and inference tokens, with a 99.5% uptime SLA; it is also sold through cloud marketplaces. Premium (SSO, private VPC links, 99.9% SLA) requires a minimum spend, and Hybrid Cloud, Private Cloud and Enterprise are priced on request. The pricing page publishes no per-unit rates in its text and points to a pricing calculator.
Pricing checked on the vendor's official page on 7 November 2026. Prices change; confirm before buying.
Who it is best for
Developers who need filtered vector search with many metadata conditions, and want an open source database with a managed option on the three main clouds.
Typical use cases
- RAG retrieval that must respect access-control or category filters
- E-commerce search combining BM25 keywords, embeddings and price or stock filters
- Multi-tenant SaaS with one shared collection partitioned by tenant
- Geo-filtered recommendations using geo payload indexes
Who should look elsewhere
- Teams that want an in-process library with no server, where LanceDB or Chroma fit better.
- Workloads that need many index algorithms such as IVF, DiskANN or GPU indexes, which Milvus documents.
Verdict
Qdrant is a focused, open source vector database whose filterable HNSW index and payload indexes make it a good fit for search that mixes vectors with many conditions. It is easy to self-host and offers several managed models, but buyers should check the calculator for Standard pricing and plan payload indexes before loading data.
Frequently asked questions
Is Qdrant open source?
Yes. The qdrant/qdrant repository is under the Apache License 2.0. Qdrant Cloud, Hybrid Cloud and Private Cloud are commercial services around the same engine.
What does the Qdrant Cloud free tier include?
A single-node cluster with 0.5 vCPU, 1 GB RAM and 4 GB disk, roughly 1 million 768-dimension vectors, with no high availability. Unused free clusters are suspended after 1 week and deleted after 4 weeks of inactivity.
Does Qdrant support hybrid search?
Yes. The Query API prefetches results from several vectors, for example BM25 sparse and dense, and fuses them with RRF or DBSF. BM25 conversion happens on the server since version 1.15.2.
How should I model multiple tenants in Qdrant?
Qdrant recommends one collection with a tenant payload field indexed with is_tenant for many small tenants, user-defined sharding for a few large tenants, or tiered multitenancy that promotes large tenants to dedicated shards.
Sources
- Qdrant pricing
- Qdrant indexing documentation
- Qdrant hybrid queries
- Qdrant multitenancy guide
- Qdrant Cloud: create a cluster (free tier limits)
- Qdrant 1.19 release post
- qdrant/qdrant GitHub repository (licence)
Checked November 2026.
Listed in: Vector Databases & AI Search.
How we research reviews and what star ratings mean: our editorial method.