- Pick it if you are:
- Teams already on Postgres; Startups building RAG prototypes
- Top complaint:
- HNSW index build is very slow once the graph outgrows memory
- Price:
- Free plan, paid from $25/mo
- Pick it if you are:
- Teams needing hybrid search; Companies wanting open-source flexibility
- Top complaint:
- Performance degrades at scale (18M+ objects)
- Price:
- Free plan, paid from $45/mo
Score Comparison
Signals Comparison
pgvector
- pgvector error: extension "vector" is not available(6/10)
Weaviate
- Steep learning curve for advanced usage(6/10)
- Documentation gaps for AI API access(4/10)
pgvector
- HNSW index build is very slow once the graph outgrows memory(7/10)
- Filtered vector search returns fewer rows than LIMIT(7/10)
- Queries do a sequential scan and ignore the vector index(6/10)
- Recall drops and memory grows at tens of millions of vectors(7/10)
Weaviate
- Performance degrades at scale (18M+ objects)(8/10)
- Memory usage grows heavily with data(7/10)
- gRPC timeout errors during batch imports(5/10)
pgvector
- operator does not exist: vector <=> unknown or real[](6/10)
- Vectors over 2,000 dimensions cannot be indexed(7/10)
- Dimension mismatch errors when you change embedding model(6/10)
- IVFFlat gives poor results when built on little data(5/10)
Weaviate
- Reindexing required when adding modules(6/10)
- Cloud console is 'incredibly barebones'(5/10)
pgvector
- HNSW build fails on shared memory or runs out of RAM(7/10)
Weaviate
- Data corruption rendered database unusable(9/10)
pgvector
- Extension availability and version lag differ by managed host(5/10)
Weaviate
- Integration challenges with existing systems(5/10)
pgvector
No signals
Weaviate
- Infrastructure costs escalate with dataset size(5/10)
Pricing Comparison
pgvector
Users: Unlimited
Storage: Your own disk (non-partitioned table up to 32 TB in Postgres)
Limits: No managed backups or support SLA; community support; 2,000-dim index limit on vector
Users: 2 active projects
Storage: 500 MB database
Limits: 500 MB database and 500 MB RAM leave little room for large HNSW indexes, pausing after a week idle, 2 projects max
Storage: 8 GB per project, then $0.125/GB
Limits: Larger RAM tiers are needed for big vector indexes
Users: 100 projects
Storage: 1 GB per project, 20 GB total
Limits: 1 GB storage per project, scale to zero after 5 minutes
Storage: Billed $0.35 per GB-month
Limits: No minimum but costs grow with CU-hours
Limits: Version depends on engine minor version
Weaviate
Users: Unlimited
Storage: Limited by your infrastructure
Limits: No managed operations, manual scaling, requires DevOps expertise
Users: 1 cluster per user
Storage: 100K objects, 1 GB memory, 10 GB disk
Limits: Best-effort availability, no backups or HA, 2 AWS regions
Users: Unlimited
Storage: Unlimited objects
Limits: Shared cluster
Users: Unlimited
Storage: Committed capacity
Limits: Prepaid commitment
Feature Comparison
| Feature | pgvector | Weaviate |
|---|---|---|
| Indexing | ||
| HNSW index | ||
| IVFFlat index | ||
| Indexing vectors above 2,000 dimensions Use halfvec (4,000) or reduce dimensions | ||
| core | ||
| Hybrid Search (Vector + BM25) | ||
| Multi-Modal Support | ||
| Cross-References | ||
| integration | ||
| GraphQL API | ||
| Kubernetes Native | ||
| Search | ||
| Cosine, L2, inner product, L1 distance | ||
| Iterative index scans for filters hnsw.iterative_scan and ivfflat.iterative_scan | ||
| Types | ||
| halfvec, bit, sparsevec types | ||
| ai | ||
| Built-in Vectorizers | ||
| security | ||
| Multi-Tenancy | ||
| RBAC | ||
| Database | ||
| SQL joins and transactions | ||
| reliability | ||
| Backups & Replication | ||
| performance | ||
| Product Quantization | ||
Have an app idea? Our small team builds any software: web and mobile apps, games, desktop, AI tools. Price agreed together.
Talk to us