Overview
Qdrant is an open-source vector database written in Rust. Building the engine in-house buys performance and memory efficiency, and the cloud billing model is kept deliberately simple.
The trade-off is explicit: effort goes into retrieval quality and memory footprint rather than a long feature list.
Key Features
- Filtering and vectors together: Payload conditions can be applied as part of the search itself, filtering before scoring. Conditional similarity search is common in recommendation and permission-scoped retrieval.
- Quantisation to compress memory: Scalar, binary and asymmetric quantisation are available, with memory footprint reducible by as much as 64 times, which is the main lever for controlling cost.
- A permanent free cloud tier: 1GB of memory, 0.5 vCPU and 4GB of disk, roughly 250,000 768-dimension vectors. A long-lived free allowance is uncommon at this price point.
- Simple billing units: The Standard tier bills by vCPU, memory and storage, with RAM around $0.078/GB-hour and backups at $0.03/GB per month; query count and vector count are not billed separately.
- Customers and funding: Users include TripAdvisor, HubSpot and Canva, and the company raised a $50M Series B in March 2026.
Use Cases
- Conditional similarity search that filters by tenant or permission scope before scoring
- Tight memory budgets that need quantisation to shrink vector footprints
- Small projects and prototypes that start directly on the permanent free tier
- Teams that prefer a simple bill without a separate charge per query
Pros
- Apache 2.0 open source with no licence cost to self-host
- A permanent free cloud tier lets small projects start at zero cost
- Filtered retrieval is built in rather than layered on in the application
- Complete quantisation options give a concrete path to lower memory cost
Pricing
Free to self-host under Apache 2.0. Cloud Free is permanent with 1GB memory, 0.5 vCPU and 4GB disk. Standard bills by vCPU, memory and storage, with RAM around $0.078/GB-hour and backups at $0.03/GB per month; queries and vector counts are not billed separately. Premium has a minimum commitment and adds SSO, VPC and a 99.9% SLA.
Summary
Qdrant subtracts in two places: engine implementation and billing units. The permanent free tier makes it a default starting point for small projects, while quantisation leaves a cost escape hatch for when scale arrives. One thing to watch: memory dominates the bill, so once dimensions and vector counts climb, tuning the compression strategy stops being optional.
Version History
- $50M Series B (2026-03): Funding directed at expanding cloud services and enterprise capability