Skip to main content
AI Biz Hub

Comparison · 9 min · 7 citations

Milvus vs pgvector Cost in 2026

Milvus vs pgvector cost in 2026: managed Zilliz Cloud, self-hosted Milvus and pgvector on Postgres priced on one scenario, with rates and free tiers.

By AI Biz Hub · Published May 25, 2026 · Updated October 19, 2026

General business information, not legal, tax or financial advice. How we check our numbers.

TL;DR

At small RAG scale in 2026, pgvector usually wins total cost when you already run Postgres with memory to spare, because the marginal cost of vectors is near zero. A database added just for vectors costs more: the Embeddings DB Cost engine prices the worked scenario, 200k 1,536-dim vectors, at $125 per month on Supabase, sized so the HNSW index fits in memory. Zilliz Cloud (managed Milvus) offers a free tier of 5 GB storage plus 2.5 million vCUs a month[1], then paid serverless or dedicated.

Self-hosting Milvus is free software but needs at least 8 GB RAM for a standalone node[3], and only becomes economical against Zilliz at roughly 100M-plus vectors. For a solo founder, the order is: pgvector if Postgres is already there, Zilliz free tier if not, self-host Milvus only at large scale.

Verified as of 2026-10-18 against the official vendor pricing and documentation pages cited below.

Order the three by scale: pgvector wins at small RAG scale if Postgres is already in the stack, the Zilliz Cloud free tier (5 GB plus 2.5 million vCUs a month) covers prototyping if it is not, and self-hosting Milvus only pays off at roughly 100M-plus vectors. The headline figure most founders chase, raw managed cost, is set at small volume by free tiers and by the memory a vector index needs; the real lever is whether you already operate Postgres and how many vectors you actually have. This article prices each option on a concrete scenario with the verified rates.

1. Three deployment options, three cost shapes

  • pgvector on Postgres. The pgvector extension adds vector search to a database you may already run. Cost is the Postgres tier plus storage, and the tier is set by memory, since the HNSW index answers fast when it fits in RAM; marginal cost of vectors is near zero if Postgres with spare memory is already there.
  • Zilliz Cloud (managed Milvus). Fully managed Milvus with a free tier and usage-based paid plans. Zero operational burden, priced on compute units and storage.
  • Self-hosted Milvus. The open-source engine on your own infrastructure. Free software, but you pay for and operate the servers, with a minimum 8 GB RAM standalone footprint[3].

The cost shapes differ: pgvector is a database bill that steps up with index memory, Zilliz is metered compute plus storage, and self-host is infrastructure plus your operational time. The cheapest depends entirely on scale and on whether Postgres is already running.

2. pgvector total cost, computed

Scenario: 200,000 vectors at 1,536 dimensions, 2,400 queries per day, 1,500 ingests per day. The Embeddings DB Cost engine prices pgvector on Supabase Pro at the compute size that holds the index.

Show the recompute-verified inputs and outputs
pgvector on Supabase + storage footprint: 200k vectors, 1,536 dim. The Postgres+pgvector line ($125) and 1.43 GB footprint are the figures for this comparison; the Pinecone ($50) and Turbopuffer ($16) columns are current-rate plan minimums but model different vendors than Milvus.
Inputs
vector_count 200000
dim 1536
queries_per_day 2400
ingest_per_day 1500
Result
vendors › row 1 › vendor Pinecone
vendors › row 1 › monthly cost 50
vendors › row 1 › notes Pinecone Standard 2026-10: $16/M read units, $4/M write units, $0.33/GB-mo, $50/mo plan minimum; a query here uses about 1.43 read units (1 per GB of namespace, minimum 0.25). Small workloads can fit the free Starter plan (up to 2 GB, 1M read and 2M write units a month) or Builder ($20/mo flat, up to 10 GB, 2M read and 5M write units).
vendors › row 2 › vendor Postgres+pgvector
vendors › row 2 › monthly cost 125
vendors › row 2 › notes Supabase Pro 2026-10: $25/mo with $10 compute credit, Large compute ($110/mo, 8 GB RAM), disk above 8 GB at $0.125/GB-mo for about 2.86 GB (table plus HNSW index). Sized from Supabase's pgvector HNSW benchmarks (1536-d: Micro 15k vectors, Small 50k, Medium 100k, Large 224k, XL 500k, 2XL 1M), scaled by floats per vector and by RAM above 2XL.
vendors › row 3 › vendor LanceDB
vendors › row 3 › monthly cost 0.25
vendors › row 3 › notes LanceDB on Cloudflare R2 2026-10: $0.015/GB-mo storage, $4.50/M writes (Class A), $0.36/M reads (Class B), one request per upsert and per query. Self-hosted compute not included.
vendors › row 4 › vendor Turbopuffer
vendors › row 4 › monthly cost 16
vendors › row 4 › notes Turbopuffer Launch 2026-10: $16/mo minimum usage; $0.33/GB-mo logical storage, $2/GB written, $0.001/TB of data queried (namespace size, 1.28 GB minimum, discounted above 32 GB), $0.05/GB returned.
cheapest vendor LanceDB
cheapest monthly cost 0.25
cheapest all in vendor Turbopuffer
cheapest all in monthly cost 16
storage gb 1.43

Computed live at build time.

The engine reports the Postgres-plus-pgvector line at $125 per month: Supabase's Pro plan at $25, which includes $10 of compute credit and 8 GB of disk[4], plus Large compute at $110[7]. Large is the smallest size that fits, because Supabase's own pgvector benchmark held 100,000 1,536-dim vectors in Medium's 4 GB and 224,482 in Large's 8 GB[6]. The 1.43 GB footprint, stored once in the table and once in the index, sits inside the included disk. If you already run Postgres for application data and it has that memory to spare, the marginal cost of adding pgvector is effectively zero.

3. Zilliz Cloud managed Milvus cost

Zilliz Cloud's free tier is 5 GB of storage plus 2.5 million vCUs of compute a month, with up to 5 collections[1]. The worked scenario (200k vectors, 1.43 GB) fits inside the free tier on storage, so a prototype runs at zero cost while its monthly compute stays within the 2.5 million vCUs.

For paid usage, storage is standardized at $0.04 per GB per month from January 1, 2026 across AWS, Azure, and Google Cloud[2]. Dedicated clusters now list per-vector rates by cluster type: from $63 per million vectors a month on performance-optimized clusters, $16 on capacity-optimized and $5 on tiered storage, measured on 768-dimension vectors[1]. So storage is verifiable at $0.04 per GB-month, and dedicated compute has published reference rates, but a 1,536-dimension index still needs the calculator for your specific cluster.

4. Self-host Milvus: free software, real infrastructure

Milvus is free open-source software, but self-hosting carries infrastructure and operational cost. A standalone Milvus node needs at least 8 GB of RAM[3], and a production cluster runs many components. At small scale, a single VM with 8 GB RAM might cost $30 to $50 per month, comparable to the managed alternatives but with you carrying the operations.

Self-hosting becomes clearly economical only at large scale, around 100M-plus vectors, where it can run several times cheaper than managed Zilliz. Below that, the operational burden (deploying, monitoring, upgrading a distributed system) outweighs the saving. For a solo founder, self-hosting Milvus is rarely the right first move; it is a scale optimization, not a starting point. The Build vs Buy Decision Engine prices this operational-cost trade-off.

5. The verified comparison table

DimensionpgvectorZilliz CloudSelf-host Milvus
Model cost$125/mo on Supabase at 200k vectors; near zero on existing Postgres[4]Free tier, then metered[1]Infrastructure only (OSS free)[3]
Free tierSupabase free 500 MB[4]5 GB + 2.5M vCUs/month[1]Unlimited (you host it)
Storage rate$0.125/GB overage (Supabase)[4]$0.04/GB-mo from Jan 2026[2]Your disk cost
Compute rateIncluded in tier[4]CU-hour (calculator)[1]Min 8 GB RAM node[3]
Operational burdenLow (managed)None (managed)High (you operate)
Best at scaleSmall, Postgres already thereSmall-to-mid, no Postgres100M-plus vectors

6. Which to pick

  1. Pick pgvector if you already run Postgres with memory for the index. Marginal vector cost is near zero, and you get SQL filtering plus transactional consistency for free.
  2. Pick Zilliz Cloud if you do not run Postgres and want zero operations. The free tier (5 GB plus 2.5 million vCUs a month) covers prototyping; paid storage is $0.04 per GB-month from January 2026.
  3. Self-host Milvus only if you are at 100M-plus vectors and have the Kubernetes capacity to operate it. Below that scale, the saving does not cover the operational cost.

For the wider picture, the cheapest vector database ranking prices eight vendors at one scenario, and the Embeddings DB Cost calculator prices pgvector against the managed stores for your own vector count.

Frequently asked questions

Is Milvus or pgvector cheaper for a small RAG app?

At solo-founder scale, pgvector usually wins on total cost if you already run Postgres with memory to spare, where the marginal cost of adding vectors is near zero. A new Postgres sized for the index is not cheap: 200k 1,536-dim vectors need Supabase's Large compute, $125 a month all in. Zilliz Cloud (managed Milvus) has a free tier of 5 GB storage plus 2.5 million vCUs a month that covers prototyping, then a serverless or dedicated paid tier. Self-hosting Milvus is free software but needs at least 8 GB RAM and grows to $300 to $600 per month only at 100M-plus vectors, far above small-app scale.

What does Zilliz Cloud cost in 2026?

Zilliz Cloud has a free tier of 5 GB storage plus 2.5 million vCUs a month, with up to 5 collections. From January 1, 2026 storage is standardized at $0.04 per GB per month across AWS, Azure, and Google Cloud. Dedicated clusters list per-vector rates by cluster type, from $63 per million vectors a month (performance-optimized) to $16 (capacity-optimized) and $5 (tiered storage), measured on 768-dimension vectors; size other dimensions in Zilliz's calculator (checked 2026-10-18).

Is self-hosting Milvus worth it?

Only at scale. Milvus is free open-source software, but a standalone deployment needs at least 8 GB RAM, and a production cluster runs many components. Self-hosting becomes economical against managed Zilliz at roughly 100M-plus vectors, where it can be several times cheaper, but at solo-founder scale the operational burden outweighs the saving and pgvector or the Zilliz free tier is simpler.

References

Sources

Primary sources only. No vendor-marketing blogs or aggregated secondary claims.

  1. 1 Zilliz — Pricing (free tier: 5 GB storage, 2.5M vCUs a month, up to 5 collections; dedicated per-vector rates by cluster type) — accessed 2026-10-18
  2. 2 Zilliz — October 2025 update (storage standardized at $0.04/GB-mo from Jan 1, 2026) — accessed 2026-10-18
  3. 3 Milvus — Requirements for Installing Milvus Standalone (minimum RAM and CPU) — accessed 2026-10-18
  4. 4 Supabase — Pricing (Pro $25/month with $10 of compute credit, enough for one Micro instance; 8 GB disk included, then $0.125 per GB) — accessed 2026-10-19
  5. 5 AI Biz Hub — Embeddings DB Cost methodology — accessed 2026-10-19
  6. 6 Supabase docs — Choosing your compute add-on (pgvector HNSW benchmark, 1,536 dimensions: Medium held 100,000 vectors in 4 GB, Large 224,482 in 8 GB) — accessed 2026-10-19
  7. 7 Supabase docs — Compute and disk (Medium $60, Large $110 a month) — accessed 2026-10-19

The 6 vendor figures on this page, each with its source and the date it was checked, are also in the vendor price reference, where every figure has its own link.

Tools referenced in this article

Related articles