GILD verdict
Start with pgvector when your data already lives in Postgres and filters are simple, so there is no second store to keep in sync.
Choose Qdrant when selective filters drive retrieval and you can index them before ingesting, and plan its memory, disk and upgrades. NCSA's Center for AI Innovation moved Illinois Chat from Qdrant to pgvector in August 2026 to cut cost and upkeep on one VM, while Malt chose Qdrant over pgvector after its own August 2023 benchmark.
pgvector stores vectors in Postgres beside the rest of your data, so there is no second store to run or sync. Qdrant runs as its own database, self-hosted as a client-server engine or on Qdrant Cloud, and builds filters into its search index. pgvector is free under the PostgreSQL License, provided the copyright notice and license paragraphs appear in all copies, while Qdrant Cloud adds a free single node and usage-based Standard pricing to the Apache-licensed engine.
Choosing between Qdrant and pgvector
Choose Qdrant if
- Queries combine vectors with selective filters you can index before ingesting, which Qdrant builds into its filterable HNSW index.
- You want built-in quantization and a managed cloud from the same vendor, with a free single node to start.
Choose pgvector if
- Your application data already lives in Postgres and you want vectors in the same tables, transactions and backups.
- You want to add an extension to the database you run, with no new service and no licence fee.
Side by side
| Dimension | Qdrant | pgvector |
|---|---|---|
| Where the vectors live | Qdrant Server uses a client-server architecture. For production the docs list Qdrant Cloud, fully managed or as Hybrid Cloud in your infrastructure, and recommend the Qdrant Private Cloud Enterprise Operator on Kubernetes to run without any cloud connection. For testing or development setups it runs as a container or binary. Qdrant Edge (Beta), built for edge devices and resource-constrained environments, runs inside the application process and can sync its local data with a Qdrant server.123 | pgvector stores vectors in Postgres beside the rest of your data.8 |
| How you run it | Qdrant runs as its own database, open-source or on Qdrant Cloud (managed, hybrid or private).34 | pgvector runs inside Postgres, self-installed or on hosted Postgres providers.8 |
| Filtered vector search | Qdrant is built for extended filtering and recommends payload indexes on every filtered field. Qdrant builds a filterable HNSW index with filter-aware edges when payload indexes exist before ingestion.256 | pgvector applies filters after the approximate index scan. From 0.8.0, iterative index scans fetch more rows when filters remove results.8 |
| Compression | Qdrant offers quantization methods including scalar (4x) and binary (up to 32x) compression.7 | pgvector offers half-precision vectors and indexes and binary quantization through expression indexes.8 |
| What you payNo cost at a stated workload: Qdrant's stored pricing pages print no per-unit rate for Standard, and pgvector's cost is the Postgres you run it on, priced by each host. | Qdrant Cloud has a free-forever single-node tier (0.5 vCPU, 1GB RAM, 4 GB disk) and usage-based Standard pricing.3 | pgvector is licensed under the PostgreSQL license, which grants use, copying, modification and distribution for any purpose without fee and without a written agreement, provided that the copyright notice and the license paragraphs appear in all copies.910 |
What practitioners report, side by side
Each product's figures come from its own practitioners, for the group you pick.
Overall signal
Teams and leaders rate pgvector −29, 11 points above Qdrant at −40. Most of the gap is in complaints, with 23 of Qdrant's 41 sources negative against 11 of pgvector's 24.
The score counts the authors from teams and leaders that take a side, one vote each with their latest view, and is how far positive votes lead negative ones, per 100 votes, so Qdrant's −40 rests on votes split 7 positive, 10 mixed and 23 negative and pgvector's −29 on 4, 7 and 10.
A lead of a fifth of the votes reads positive and a lead of half strong positive, with the same lines below zero. The dots show each source, positive, mixed, negative. Run a pilot on your own documents and filters to measure recall against exact search, query latency at your filter selectivity, and memory per million vectors.
Footprint on the AI Application Platform stack
Both cover the same 9 layers at the same depth. Coverage is each product's placement on this stack, as its product page shows it. Core is a job the product owns; Supports and Touches are jobs it helps with. Coverage says nothing about how well a product does them.
| Stage | Qdrant | pgvector |
|---|---|---|
| Model2 layers | Touches 2 | Touches 2 |
| Build2 layers | Touches 2 | Touches 2 |
| Data1 layer | Core 1 | Core 1 |
| Evaluate1 layer | Touches 1 | Touches 1 |
| Operate3 layers | Supports 1, Touches 2 | Supports 1, Touches 2 |
Show all 9 layers
| Layer | Qdrant | pgvector |
|---|---|---|
| Model | ||
| Model APIs | Touches | Touches |
| Model gateway | Touches | Touches |
| Build | ||
| Orchestration | Touches | Touches |
| Tool use | Touches | Touches |
| Data | ||
| Retrieval | Core | Core |
| Evaluate | ||
| Evaluation | Touches | Touches |
| Operate | ||
| Observability | Touches | Touches |
| Cloud deployment | Supports | Supports |
| Guardrails | Touches | Touches |
What practitioners praise
Among teams and leaders, Low Operational Work reaches 5 sources from 3 authors only for Qdrant, at 6 of 7. pgvector needs a pilot on the same retrieval queries before you choose on it.
| Praised for | Qdrant | pgvector |
|---|---|---|
| Low Operational Work | 6 of 7 | Not a theme in its sources |
Each count is the sources that praise a theme, or raise it as a concern, out of those that discuss it. A row pairs the two products when both carry a theme under the same or an equivalent name.
A count ending in "so far" has under 5 sources from 3 authors among teams and leaders. "Not a theme in its sources" means GILD found no theme of that name in that product's sources.
What practitioners raise as concerns
Among teams and leaders, Qdrant alone has four concern themes that reach 5 sources from 3 authors. Cluster Operations and Memory and Disk Sizing lead, at 9 of 9 and 7 of 7. Filters and the Query Planner and Releases and Host Versions reach 5 sources from 3 authors only for pgvector, at 4 of 5 each. Qdrant needs a pilot on the same retrieval queries before you choose on either.
| Raised as a concern | Qdrant | pgvector |
|---|---|---|
| Cluster Operations | 9 of 9 | Not a theme in its sources |
| Memory and Disk Sizing | 7 of 7 | Not a theme in its sources |
| Upgrade Regressions | 6 of 6 | Not a theme in its sources |
| A Separate Server to Run | 5 of 5 | Not a theme in its sources |
| Filters and the Query Planner | Not a theme in its sources | 4 of 5 |
| Releases and Host Versions | Not a theme in its sources | 4 of 5 |
12 sources
- Qdrant docs: Installation (Markdown)
- Qdrant on GitHub: README
- Qdrant: Pricing
- Qdrant docs: Overview
- Qdrant docs: Filtering
- Qdrant docs: Indexing
- Qdrant docs: Quantization
- pgvector on GitHub: README
- pgvector on GitHub: LICENSE
- pgvector on GitHub: META.json
- Center-for-AI-Innovation/Illinois-Chat issue #97: replace Qdrant with pgvector (Jul 27, 2026)
- Malt engineering blog: a vector database for freelancer recommendation (Jul 10, 2024)