Index › Data platform › Vector databases › pgvector vs Milvus

Vector databases · October 2026 Edition

pgvector vs Milvus

Eight of fourteen models named pgvector first on the direct prompt; zero named Milvus. Both were named by all fourteen models and pgvector carries 66 labels and Milvus 41, so the shares are not directly comparable.

pgvector

endorsed leader

Named in one category this edition.

Milvus

accepted challenger

Named in one category this edition.

First-choice share35%3%Of first choices across the direct, paraphrase, budget and scale prompts, 0 to 100.

Negative rate6%20%Negative labels as a share of the product's labels, 0 to 100.

Rank in category#1#6A position in a field of 10; printed, not drawn.

Labels6641A count; the two differ.

The two percentage rows are drawn on one 0 to 100 track, pgvector reading right to left. Rank and label count are printed, not drawn.Weaviate was named alongside these two in eleven of the fourteen direct answers. pgvector vs Pinecone · pgvector vs Qdrant · pgvector vs Chroma

Share is the count of first choices across the direct, paraphrase, budget and scale prompts over all fourteen models, for a mid-market B2B company; rank is within the category; every quote names the model and the prompt it came from. Both figures come from the vector databases page.

By framing

How many of the fourteen models made each the first choice, per way of asking, and how many argued against it.

pgvectorFirst choices, of fourteen modelsMilvus

Direct801 against Milvus

Paraphrase30

Comparative111 against Milvus

Budget-constrained823 against Milvus

Scale-constrained40

Negative704 against pgvector · 3 against Milvus

Bars are first choices, 0 to 14 each sideModels that argued againstA model can name both, so the two sides of a row do not sum to fourteen.

Across every category in the October 2026 Edition, pgvector and Milvus were named in the same answer ninety-seven times, of the 179 answers naming pgvector and the 140 naming Milvus. In those answers Milvus took the first choice four times and pgvector thirty-seven.

Every model, every framing

The eighty-four answers behind the chart above, one cell each: where pgvector and Milvus stood in it.

Model	DirectPG	ParaphrasePG	ComparativePG	Budget-constrainedPG	Scale-constrainedPG	NegativePG
Claude Haiku 4.5		PG	PG			PG
GPT-5.4 mini			PG	PG
Gemini 3.5 Flash	PG	PG	PG	PG	PG	PG
Perplexity Sonar	PG	PG	PG	PG
Grok 4.1 Fast	PG	PG	PG	PG	PG	PG
Mistral Small	PG			PG		PG
DeepSeek V4 Flash	PG	PG	PG	PG	PG	PG
Llama 4 Maverick	PG			PG		PG
Qwen 3.7 Flash				PG	PG	PG
Kimi K2	PG		PG	PG	PG	PG
GLM 4.7 FlashX	PG	PG	PG	PG		PG
MiniMax M2.5		PG	PG	PG		PG
GPT-6 Luna	PG	PG	PG	PG	PG	PG
Muse Glimmer 30B	PG			PG	PG

PG pgvector

MilvusPG first choicePG named as an alternativePG argued againstblank: not namedEach cell is one answer, pgvector on the left and Milvus on the right.

The direct prompt

The plain question, one answer per model, grouped by where pgvector and Milvus stood in it.

pgvector first, Milvus not the choice

8 of 14 modelsMilvus was named in the answer but not as the choice, or not at all.

Gemini 3.5 Flashpgvector alternatives: Pinecone, Qdrant, Weaviate

Perplexity SonarQdrant, pgvector alternatives: Pinecone, Weaviate

Grok 4.1 Fastpgvector alternatives: Pinecone, Qdrant, Weaviate

Mistral Smallpgvector alternatives: pgvectorscale

DeepSeek V4 FlashQdrant, pgvector alternatives: Pinecone, Weaviate

Llama 4 Maverickpgvector

Kimi K2pgvector alternatives: Pinecone, Qdrant, Weaviate

Muse Glimmer 30BPinecone, pgvector alternatives: Qdrant, Weaviate

Neither was the first choice, one was named

3 of 14 modelsThe answer put something else first and named one of the two as an alternative.

GLM 4.7 FlashXQdrant, Weaviate alternatives: pgvector

MiniMax M2.5Pinecone alternatives: Elasticsearch, Milvus, Weaviate

GPT-6 LunaQdrant alternatives: Pinecone, pgvector

Neither was named

3 of 14 modelsThe answer made no first choice from these two in this category.

Claude Haiku 4.5Pinecone, Weaviate

GPT-5.4 miniPinecone alternatives: Qdrant, Weaviate

Qwen 3.7 FlashPinecone alternatives: MongoDB Atlas Vector Search, Weaviate

Bold names in an answer are the products the judge labeled a first choice; a model naming several gives each of them that label. The full answer text for every row is in the record.

By buyer segment

The same question asked on behalf of a different buyer. Each standing is computed within its segment and they are never added together. The figures above are the mid-market standing, which is the one the category orders by.

Small business

pgvector leads by forty-nine points.

pgvector#1 of 7

Milvus#7 of 7

The full small business standing →

Mid-marketThe figures above

pgvector leads by thirty-two points.

pgvector#1 of 10

Milvus#6 of 10

The full mid-market standing →

Enterprise

The order flips: Milvus leads at enterprise.

Milvus#3 of 10

pgvector#5 of 10

The full enterprise standing →

What the models said about pgvector

Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Six of eight in this category shown.

“| Poor filtered search | pgvector, basic setups | ... | Reindexing locks | pgvector | Large HNSW rebuilds block reads/writes” Grok 4.1 Fast · negative prompt · soft negative

“pgvector when you expect very high vector-search throughput or huge scale without testing” GPT-6 Luna · negative prompt · soft negative

“Be cautious at massive scale ... you must be cautious of its ceilings” Gemini 3.5 Flash · negative prompt · soft negative

“Unless you have 2-3 people who can own this infrastructure, lean managed or existing DB extension first.... Already run Postgres heavily, <10M vectors | pgvector” Kimi K2 · scale prompt · first choice

“The best vector database for a mid-market B2B company is pgvector, which is a PostgreSQL extension. It is the strongest default for mid-market enterprises.” Llama 4 Maverick · direct prompt · first choice

“evaluating pgvector might save you the overhead of learning a new technology, as it handles vector search well within a relational environment” Qwen 3.7 Flash · scale prompt · first choice

What the models said about Milvus

Every negative label with a quote, up to three, then the highest-weighted positives, up to three. Six of eight in this category shown.

“heavy self-hosted systems like Milvus (requires Kubernetes know-how) ... neither is friendly to a limited budget” DeepSeek V4 Flash · budget prompt · hard negative

“Caution: If you run Milvus self-hosted, you must be on 2.4.24+, 2.5.21+ or 2.6.5+ and strip `sourceID` at the gateway” Muse Glimmer 30B · negative prompt · soft negative

“handles billions of vectors at lower cost; requires engineering resources... more resource-intensive to operate” Claude Haiku 4.5 · comparative prompt · soft negative

“likely to be an open-source option such as Milvus or Chroma, as they offer free or low-cost options” Llama 4 Maverick · budget prompt · first choice

“I'd recommend starting with either self-hosted Milvus/Qdrant (if you have DevOps capacity)” Claude Haiku 4.5 · budget prompt · first choice

“Massive Scale / Billions of vectors | Milvus” Qwen 3.7 Flash · comparative prompt · first choice

Also compared

pgvector vs Pinecone pgvector vs Qdrant pgvector vs Chroma pgvector vs Weaviate pgvector vs MongoDB Atlas Vector Search Pinecone vs Milvus Qdrant vs Milvus Chroma vs Milvus Weaviate vs Milvus Milvus vs MongoDB Atlas Vector Search All twenty-one head-to-head pages and the full vector databases standing

Comparisons are drawn for the top eight products in each category, each against each. The output is the models' output; nothing here is a recommendation by the index.