Qdrant review, pricing and verdict
Open-source vector search engine with a managed cloud on AWS, GCP and Azure
- Best for
- ML engineer
- Not for
- Non-technical founder who wants a ready-made search box
Should you use Qdrant?
Pick Qdrant when you want vector search you can start on a free cluster and later move to managed cloud or your own servers without changing engines. Skip it if you need a published rate card or a serverless product today. More than 100 staff work on it, and release 1.20 shipped on 22 October 2026.
One Rust engine sold as open source, managed cloud on three providers, Hybrid Cloud and Private Cloud.
The engine source is public on GitHub, so a team can move from the managed service to self-hosting, which caps how far lock-in to Qdrant Cloud can go.
Also for
- Backend engineer
- ML and backend engineers building retrieval for RAG and search
Show 2 more
- Platform teams that need managed clusters inside their own cloud account
- Startups prototyping on a free vector database before paying
Main trade-off
The Standard tier has no published rate card, so cost needs the calculator.
Qdrant pricing
Server plans
| Plan | CPU | RAM | Storage | Price |
|---|---|---|---|---|
| Free TierCloud · Testing and prototypes · Limited choice of cloud providers and regions | 0.5 vCPU | 1 GB | 4 GB | $0/mo |
| Open source (self-hosted)Cloud · Teams that run the engine themselves · Your own infrastructure | — | — | — | $0/mo |
| Standard TierCloud · Production workloads and scaling applications · AWS · GCP · Azure | — | — | — | — |
| Premium TierCloud · Enterprises with extra security and compliance needs · AWS · GCP · Azure | — | — | — | — |
| Hybrid CloudCloud · Local data residency and regulated workloads · Your own infrastructure | — | — | — | — |
| Private CloudCloud · Large enterprises and air-gapped setups · Your own infrastructure | — | — | — | — |
- Minimum commitment
- None on Free and Standard; Premium has a minimum spend
- Billing
- Pay as you go · Hourly · Freemium
The Free tier costs nothing and needs no credit card. Standard charges by consumption each hour across cluster compute, memory and storage, plus backups and paid inference tokens; Qdrant shows no rate card, only a pricing calculator. Premium needs a minimum spend, and Hybrid Cloud and Private Cloud are quoted by the sales team. The self-hosted engine is free software and you pay only for your own servers.
Costs that are easy to miss
- Billing covers backup storage and paid inference tokens as well as vCPU, memory and disk (Qdrant pricing FAQ).
- Multi-AZ clusters are billed at a higher rate to cover cross-zone traffic (Qdrant cluster docs).
Qdrant regions and compliance
- Uptime SLA
- 99.5% Standard, 99.9% Premium
Available regions
- Multiple regions on AWS
- Multiple regions on GCP
- Multiple regions on Azure
- Your own infrastructure (Hybrid Cloud)
Compliance certifications
- SOC 2 Type II
- HIPAA
- GDPR
As stated by the provider. Ask for the current report before you rely on one.
Managed services
- Managed vector database
- Hybrid Cloud
- Private Cloud
- Cloud Inference
- Backup and disaster recovery
Access methods
- Web console
- API
- CLI
- Dashboard
Networking
- Private networksSupported
Storage services
- Backup and snapshotsSupported
What Qdrant does
Qdrant is an open-source vector search engine built in Rust by a Berlin company. It stores embeddings and returns the nearest matches for retrieval, recommendations and agent memory. You can self-host it or buy the managed Qdrant Cloud across AWS, GCP and Azure, plus Hybrid and Private Cloud. Its repository has 29,000 GitHub stars.
Open-source engine in Rust
The engine is public software with 29,000 GitHub stars and more than 250 million downloads across its packages, according to Qdrant's about page.
Managed cloud on three providers
The managed service deploys to AWS, GCP and Azure, and clusters can be driven through the API, Terraform, Pulumi and a CLI.
Hybrid and Private Cloud
Managed clusters can run on your own compute, network and storage so data stays inside your network.
BM25 on full-text indexes
BM25 scoring on full-text indexes arrived in 1.20 and replaces a separate sparse-vector setup for keyword ranking.
Show all 7 features
Quantisation options
The newest release adds turbo16, turbo8 and bits8 datatypes to cut the memory vectors take.
GPU-accelerated indexing
Qdrant says GPU indexing builds an index up to 4x faster for bulk loads, available on AWS clusters.
Security controls
Traffic is encrypted in transit and at rest, and Premium adds SSO, private links and customer-held encryption keys.
Strengths and limits
Strengths
- The free cluster needs no credit card and holds about 1 million 768-dimension vectors in a single small node, enough to prototype retrieval before any spend.
- One open-source engine covers self-hosting, Hybrid Cloud and the hosted tier, and a migration tool from open source to Cloud means a move keeps the same engine.
- SOC 2 Type 2 and HIPAA attestations plus a GDPR data processing agreement give a security review documents to read, available through its Trust Center.
- Customers named on its Series B post include Canva, HubSpot, Roche, Bosch and OpenTable, which shows production use at large teams.
Limits
- A single node typically tops out around 100 million vectors, fewer for high-dimension float32 data (Qdrant capacity docs, read 15 November 2026), so large collections need multi-node clusters, with at least 3 nodes advised for production.
- The pricing page gives no per-unit rates for the Standard tier; cost is estimated in a calculator, so a team cannot set a budget from public pages alone.
- An idle free cluster is paused after seven days and removed three weeks after the notice, per Qdrant's billing docs.
- Teams that pay through the AWS, GCP or Azure marketplace can create paid clusters only in that provider's regions (Qdrant billing docs, read 15 November 2026).
Qdrant alternatives
| Service | Regions | Pricing | Compliance | Checked |
|---|---|---|---|---|
| Qdrant This page | -- | Free tier · Pay as you go · Hourly | SOC 2 · HIPAA · GDPR | 15 Nov |
 PineconePick Qdrant if you want an engine you can also run on your own servers; pick Pinecone if you prefer its flat-fee Builder plan at entry level. | US · Europe · Asia Pacific | Free tier · Pay as you go · Subscription | SOC 2 · GDPR · ISO 27001 +1 | 13 Nov |
Qdrant facts
- Headquarters
- Berlin, Germany
- Total funding
- Series B led by AVP, announced 12 March 2026
- Team size
- 100+
- Open source
- Yes
What changed
- Release Qdrant 1.20: full-text search improvements source
Questions about Qdrant
How much does Qdrant Cloud cost?
The Free tier costs nothing. Standard is billed by usage, hourly, on the vCPU, memory and storage of your clusters, plus backup storage and paid inference tokens, and Qdrant publishes no rate card, only a calculator. Marketplace payers see usage as $0.01 Resource Usage Units, so a cluster costing $85 in a month adds 8,500 units to the cloud bill. Premium needs a minimum spend and Hybrid and Private Cloud are quoted.
Is there a free Qdrant tier?
Yes. The Free tier is a single-node cluster with 0.5 vCPU, 1 GB RAM and 4 GB disk, and it needs no credit card. Qdrant says it serves about 1 million vectors of 768 dimensions. A free cluster is suspended after one week without use and deleted three weeks after the suspension notice.
Which clouds and regions does Qdrant Cloud support?
Qdrant Cloud runs on AWS, GCP and Azure, and Hybrid Cloud runs on your own infrastructure. The current region list appears in the cluster creation flow and on the pricing page, and new regions are added when customers ask. If you pay through a cloud marketplace, paid clusters are limited to that provider's regions.
Does Qdrant support hybrid keyword and vector search?
Yes. Since the 1.20 release on 22 October 2026, BM25 scoring runs on full-text indexes, so keyword ranking works without building sparse vectors. The release notes also describe a migration path for teams that used sparse BM25 before. A new substring match condition arrived with it.
Does Qdrant meet SOC 2 and HIPAA requirements?
Qdrant states that it holds SOC 2 Type 2 and HIPAA certifications and publishes the reports in its Trust Center. For GDPR it provides a data processing agreement. Contact Qdrant before storing protected health information, since handling depends on your deployment.
About Qdrant
Qdrant is a Rust-built vector search engine from Qdrant Solutions GmbH, a company based in Berlin. Its founders, André Zayarni and Andrey Vasnetsov, started the work in 2021 after finding that libraries such as FAISS did not give them the features and scale a matching engine for unstructured data needed. The company announced a $50 million Series B on 12 March 2026, led by AVP. The engine is open source and also sold as Qdrant Cloud, a managed service on AWS, GCP and Azure, with Hybrid Cloud and Private Cloud for teams that must keep data inside their own network. Its job is nearest-neighbour search over embeddings, the layer that retrieval for RAG, recommendations and agent memory sits on, and it belongs to the vector database section of the directory.
Read the full background
What sets it apart from Pinecone is that the engine itself is public software: the same code runs on your laptop, on your own servers, inside Hybrid Cloud and in the managed service, and the pricing page lists the open-source engine as its own column beside the cloud tiers. Version 1.20, released on 22 October 2026, added BM25 scoring on full-text indexes without sparse vectors, 16-bit and 8-bit TurboQuant datatypes and internal authentication by default on clusters. Another vector store listed on Hokai is LanceDB, and G2 lists Weaviate and Pinecone among the products it compares with Qdrant.
Qdrant fits engineers who build retrieval for assistants and search, platform teams that want managed clusters running in their own cloud account, and startups that want to prototype on a free cluster before they pay. Frameworks such as LangChain and LlamaIndex sit above a store like this, and the guide to AI search and retrieval infrastructure weighs it against the other options. Its own site names Canva, HubSpot, Roche, Bosch and OpenTable among customers in production.
It is a poor match for a team that wants a published rate card before talking to anyone, because the Standard tier is billed by usage and sized in a calculator. It is also the wrong pick for anyone who needs a fully serverless product today: Qdrant Serverless is a waitlist and Qdrant Edge is in beta. The call: choose Qdrant if you want to start on the free cluster and keep the option of running the engine yourself, and choose Pinecone if you want one managed service with the least operational work.
Sources and guides
Official sources
- Qdrant official site Vendor
- GitHub repository GitHub
Plans, regions, the uptime commitment and the compliance list on this page are read from Qdrant's own pages and documentation. The verdict is Hokai's reading of them, not the provider's claim.