Vector Database Pricing: Pinecone vs Weaviate vs Qdrant vs pgvector (September 2026)

Q: How much does it cost to store embeddings?

Storage cost = vectors x dimensions x 4 bytes per vector (float32). For 1 million 1,536-dimension vectors: 1M x 1,536 x 4 = 5.72 GB. At Pinecone serverless rates ($0.33/GB/month), that is $1.89/month. pgvector on an existing Postgres database is near-zero marginal cost.

Q: How much does Pinecone cost for 1 million vectors?

For 1 million 1,536-dimension float32 vectors: storage is 5.72 GB at $0.33/GB = $1.89/month on Pinecone serverless. Add read and write unit charges on top. The free Starter plan covers up to 2 GB; the Standard plan carries a $50/month minimum. Pod-based indexes are legacy and unavailable to Standard or Enterprise sign-ups from 18 August 2025. On-demand serverless caches data best-effort, so cold queries fetch from object storage; Pinecone says dedicated read nodes eliminate cold starts.

Q: What are the limits of pgvector?

pgvector works well up to approximately 10-50 million vectors depending on dimension count and RAM available. Beyond that, IVFFlat and HNSW index build times become very long and query latency increases. At 100M+ vectors, managed vector databases (Qdrant, Weaviate, Zilliz) are significantly better. For small applications, pgvector is excellent and near-zero cost on an existing Postgres instance.

Q: What is a Pinecone serverless cold query?

On Pinecone's default on-demand serverless capacity, caching is best-effort: frequently accessed data is cached, but cold queries fetch from object storage. Pinecone says its dedicated read nodes, billed at fixed hourly rates for provisioned hardware, keep all index data warm in memory and on local SSDs and eliminate cold starts. Pinecone does not publish a cold-query latency figure on that page.

Q: Is pgvector on Supabase free?

Supabase's free tier includes pgvector with 500MB database storage, which can hold approximately 85,000 1,536-dimension float32 vectors. For larger corpora, Supabase Pro starts at $25/month with 8GB storage. pgvector on a self-hosted Postgres instance has near-zero marginal cost beyond the server.

Q: What is the cheapest vector database?

pgvector on an existing Postgres instance is the cheapest at near-zero marginal cost. Among managed options, Weaviate Cloud publishes a Flex storage rate from $0.12/GiB but also bills per vector dimension, with a $45/month Flex minimum. Qdrant Cloud has a free 1GB RAM / 4GB disk cluster and bills paid clusters on vCPU, RAM and disk. Pinecone serverless storage is $0.33/GB, with a free Starter plan up to 2 GB. For pure storage cost minimization, run Qdrant self-hosted on a small EC2 instance at about $0.023/GB.

Embedding generation is a one-time cost. Storage is forever. This page covers the storage formula, the major vector database options at real-world scales, and the dimension-reduction savings nobody mentions.

Verified September 2026

Storage Cost Calculator

Vector count

Dimensions

Binary (4x less storage)

1,000,000 vectors x 1536 dims x 4 bytes = 5.7 GB

Pinecone Serverless

Free Starter plan up to 2 GB

$1.89/mo

Qdrant Cloud

Est.; free tier 1GB RAM / 4GB disk

$0.69/mo

Weaviate Cloud

Flex storage rate; plus vector dims

$0.69/mo

Zilliz Cloud

Best for 1B+ scale

$0.57/mo

pgvector (self-hosted)

Near-zero on existing DB

$0.13/mo

Storage-footprint estimate only. Pinecone serverless and Weaviate Cloud (Flex) publish per-GB storage rates, but Weaviate also bills per vector dimension with a $45/month Flex minimum. Qdrant and Zilliz bill on provisioned RAM and compute, so their real cluster cost is higher than the per-GB figure shown here. Add query/read costs on top. Binary encoding requires Titan V2 (AWS) or custom quantization.

The Storage Formula

Each embedding vector is stored as 32-bit floats. The raw storage size before index overhead:

bytes = vectors x dimensions x 4
# Example: 1M vectors at 1536 dims
= 1,000,000 x 1,536 x 4 = 5.72 GB
# Add 20-40% for HNSW index overhead
effective = ~7-8 GB total

Managed vector databases charge for both the raw vectors and the index. HNSW index adds 20-40% overhead on top of raw vector storage. For pricing purposes, most providers bill for total storage used, including index. The calculator above uses raw vector math; add ~30% for real-world managed DB estimates.

Monthly Storage Cost by Scale

Scale	Raw size	Pinecone Serverless	Qdrant Cloud (est.)	Weaviate Cloud	Zilliz Cloud	pgvector
1M vectors (1536d)	5.7 GB	$2/mo	$0.69/mo	$0.69/mo	$0.57/mo	$0.13/mo
10M vectors (1536d)	57.2 GB	$19/mo	$7/mo	$7/mo	$6/mo	$1/mo
100M vectors (1536d)	572.2 GB	$189/mo	$69/mo	$69/mo	$57/mo	$13/mo

Raw float32 storage. Add 20-40% for HNSW index overhead in practice. Query costs not included. Pinecone serverless also charges per read/write operation.

Provider-by-Provider Analysis

Each managed vendor has a dedicated pricing guide with its published plan table, free tier and sources: Pinecone, Qdrant and Weaviate.

Pinecone

$0.33/GB/month (serverless)

Strengths

+ Purpose-built vector DB
+ Free Starter plan (up to 2 GB storage)
+ Managed indexes, no ops burden

Watch out

- Cold queries fetch from object storage (on-demand)
- Standard plan has a $50/month minimum
- More expensive per GB than alternatives

Best for: prototypes and small production apps where ops burden is unacceptable. Cold queries on on-demand capacity are the main risk for latency-sensitive applications; dedicated read nodes remove them at a fixed hourly cost.

Full Pinecone pricing guide Pricing page

pgvector

~$0.023/GB on self-hosted Postgres

Strengths

+ Near-zero marginal cost on existing DB
+ SQL-compatible; standard tooling
+ No extra infrastructure

Watch out

- Scales to ~50M vectors before quality degrades
- HNSW index build time grows with data
- No built-in filtering at extreme scale

Best for: applications already on Postgres with under 50M vectors. The default choice for small-to-medium RAG applications. Move to a dedicated vector DB at scale.

Pricing page

Qdrant

Free 1GB RAM / 4GB disk; paid billed hourly

Strengths

+ Strong filtering at scale
+ Open-source self-hosting option
+ Rust-based; fast and memory efficient

Watch out

- Less managed tooling than Pinecone
- Sizing requires RAM+disk understanding

Best for: production RAG at medium scale (5M-500M vectors), especially with metadata filtering requirements. Self-hosting on EC2 is cost-effective at 50M+ vectors.

Full Qdrant pricing guide Pricing page

Weaviate Cloud

from $0.12/GiB storage + vector dims (Flex)

Strengths

+ Free plan: 100,000 objects, no card
+ GraphQL and REST APIs
+ BM25+vector hybrid search built-in

Watch out

- Flex plan minimum is $45/month
- Bills per vector dimension as well as storage

Best for: hybrid search (keyword + semantic) use cases. The BM25 integration is first-class and often eliminates the need for a separate keyword search layer.

Full Weaviate pricing guide Pricing page

Zilliz Cloud (Milvus)

$0.10/GB/month

Strengths

+ Best for billion-scale corpora
+ Open-source Milvus-compatible
+ Strong horizontal scaling

Watch out

- Overkill for <10M vectors
- More operational complexity

Best for: billion-scale document corpora, media libraries, and enterprises with existing Milvus deployments. Comparable pricing to Qdrant with better 1B+ scale story.

Pricing page

Dimension Reduction: The Storage Shortcut

For models that support Matryoshka Representation Learning ( OpenAI text-embedding-3-large, Google Gemini models), you can request fewer dimensions via the API. Storage cost scales linearly with dimensions:

Dimensions	GB per 100M vecs	Pinecone cost/mo	Savings vs 3072d
3072 (full)	1144.4 GB	$378	baseline
1536 (half)	572.2 GB	$189	$189 saved
768 (quarter)	286.1 GB	$94	$283 saved

For 100M vectors at 100M-scale, halving dimensions saves $189/month on Pinecone with minimal quality impact. MTEB score drop: roughly 1-2 points.

Frequently Asked Questions

How much does it cost to store embeddings?

Storage cost = vectors x dimensions x 4 bytes. For 1M 1,536-dim vectors: 5.72 GB. At Pinecone serverless rates ($0.33/GB), that is $1.89/month. pgvector on an existing Postgres database is near-zero marginal cost.

How much does Pinecone cost for 1 million vectors?

For 1M 1,536-dim float32 vectors: 5.72 GB at $0.33/GB = $1.89/month storage on Pinecone serverless. Add read and write unit charges on top. The free Starter plan covers up to 2 GB; the Standard plan carries a $50/month minimum. Pod-based indexes are legacy and unavailable to Standard or Enterprise sign-ups from 18 August 2025. On-demand serverless caches data best-effort, so cold queries fetch from object storage; Pinecone says dedicated read nodes eliminate cold starts.

What are the limits of pgvector?

pgvector works well up to approximately 10-50 million vectors. Beyond that, IVFFlat and HNSW index build times become very long and query latency increases. At 100M+ vectors, managed vector databases are significantly better.

What is a Pinecone serverless cold query?

On Pinecone's default on-demand serverless capacity, caching is best-effort: frequently accessed data is cached, but cold queries fetch from object storage. Pinecone says its dedicated read nodes, billed at fixed hourly rates for provisioned hardware, keep all index data warm in memory and on local SSDs and eliminate cold starts. Pinecone does not publish a cold-query latency figure on that page.

Is pgvector on Supabase free?

Supabase's free tier includes pgvector with 500MB database storage, which can hold approximately 85,000 1,536-dimension float32 vectors. Supabase Pro starts at $25/month with 8GB storage.

What is the cheapest vector database?

pgvector on an existing Postgres instance is the cheapest at near-zero marginal cost. Among managed options, Weaviate Cloud publishes a Flex storage rate from $0.12/GiB but also bills per vector dimension, with a $45/month Flex minimum. Qdrant Cloud has a free 1GB RAM / 4GB disk cluster and bills paid clusters on vCPU, RAM and disk. Pinecone serverless storage is $0.33/GB, with a free Starter plan up to 2 GB. For pure storage cost minimization, run Qdrant self-hosted on a small EC2 instance at about $0.023/GB.

Full calculator

Storage + embedding + query cost combined

Self-hosted break-even

When to stop using managed vector DBs

Optimization tips

Dimension reduction, binary embeddings

Disclaimer: Independent resource. Pinecone serverless ($0.33/GB/month) and Weaviate Cloud Flex (from $0.12/GiB, plus a per-vector-dimension charge) publish storage rates, read from their pricing pages on 1 October 2026. Qdrant and Zilliz price on provisioned RAM and compute rather than a published per-GB rate, so the figures shown for them are storage-footprint estimates for comparison, not quoted list prices; real managed-cluster cost is higher once RAM, compute, and query load are included. Always confirm current pricing at each provider's pricing page before making architecture decisions.