IT AI Index
September 2026 Edition · The permanent record of this edition. The unqualified address always carries the latest edition.
Index › Data platform › Vector databases › Enterprise › September 2026 Edition

Vector databases for enterprise buyers

Asked as “vector database”, and as “vector search database for AI applications”, on behalf of an enterprise B2B company. 43 first choices recorded across the direct, paraphrase, budget and scale prompts, twelve models each.
Standing
Pinecone
Clear leader
53% of first choices, clear leader.
ShowinginAll 67 categories in the September 2026 Edition →

By buyer segment

The same question asked on behalf of a different buyer. Each standing is computed within its segment; they sit side by side and are never added together.
Small businesspgvector46%Clear leaderStanding →Mid-marketQdrant37%ContestedStanding →EnterprisePinecone53%Clear leaderThis page

01The standing

Share is the count of first choices across the direct, paraphrase, budget and scale prompts, over all twelve models, for an enterprise B2B company. Ordered by share.
ProductFirst-choice shareNegative rateLabelsQuadrant
01Pinecone53%28%64criticized default
02Qdrant19%2%50accepted challenger
03pgvector12%31%54criticized challenger
04Weaviate9%11%61accepted challenger
05Zilliz Cloud5%13%15accepted challenger
06Milvus / Zilliz Cloud2%0%13accepted challenger
Show the five products at 0%, ordered by negative rate
11Chroma0%67%15criticized challenger
09Elasticsearch0%20%15accepted challenger
07Milvus0%20%41accepted challenger
10Redis0%15%13accepted challenger
08Azure AI Search0%0%13accepted challenger
Bars are the share of first choices, 0 to 100Every product with at least 10 labels here. Every product name links to its vendor page.
Head to head
Qdrant vs pgvectorQdrant vs Pineconepgvector vs Pinecone

Recommended versus criticized

Every product with at least 10 labels here, on both axes. The 30% line names a quadrant, not the verdict above: that one needs more than 40%.

Criticized challengerCriticized default
Negative label rate →
01
02
03
04
05
06
07
08
09
10
11
Accepted challengerEndorsed leader
0%First-choice share → · lines at 30% share and 25% negative70%
Key
01Pinecone53%
02Qdrant19%
03pgvector12%
04Weaviate9%
05Zilliz Cloud5%
06Milvus / Zilliz Cloud2%
07Milvus0%
08Azure AI Search0%
09Elasticsearch0%
10Redis0%
11Chroma0%

02What they warned about

Six of twelve models held their first choice under the paraphrase. Gemini 3.5 Flash, Perplexity Sonar, Qwen 3.7 Flash, Kimi K2, GLM 4.7 FlashX and MiniMax M2.5 changed. A high negative share on a product with few labels is a warning. A low share on a product with many labels is salience, not sentiment.
Pinecone
28%
18 of 64 labels negative · 11 of 12 models · 4 hard negative
“Avoid "Serverless" plans (like Pinecone Serverless or standard AWS Opensearch Vector Engine) for thousands of users.” Qwen 3.7 Flash, budget prompt
pgvector
31%
17 of 54 labels negative · 8 of 12 models
“**Postgres + pgvector** if you already standardize on Postgres and want to inherit existing controls, though you may trade off scale and specialized search features” GPT-5.4 mini, scale prompt
Chroma
67%
10 of 15 labels negative · 7 of 12 models · 6 hard negative
“Many "developer‑friendly" tools (e.g., Chroma) are excellent for prototypes but lack enterprise features ... avoid them for mission‑critical RAG pipelines” GLM 4.7 FlashX, negative prompt
Milvus
20%
8 of 41 labels negative · 7 of 12 models · 1 hard negative
“avoid highly complex C++ or Python-based microservices architectures like **Milvus** or **Zilliz** (when self-hosted) unless you have a dedicated Site Reliability Engineering (SRE) team” Qwen 3.7 Flash, negative prompt

03What they cite

Citations exist only for the models that return a source list: twelve of the twelve in this edition, and all six flagship models on the expanded tier.

Sites the answers cite

67 of 72 answers in this category came back with a source list, from 12 of 12 models: citations where the model returns them, or the search results it consulted. 5 of those lists are Google grounding redirects that name no site and are left out of the counts. 851 links across 263 sites, every framing counted. Ranked by the number of answers carrying the site or page.

atlan.com
32 answers · 32 citations · 9 models
encore.dev
30 answers · 30 citations · 9 models
firecrawl.dev
30 answers · 30 citations · 10 models
iternal.ai
24 answers · 24 citations · 8 models
sganalytics.com
21 answers · 21 citations · 9 models
medium.com
20 answers · 25 citations · 9 models
aloa.co
19 answers · 20 citations · 7 models
g2.com
18 answers · 18 citations · 8 models
dev.to
17 answers · 22 citations · 7 models
actian.com
17 answers · 19 citations · 9 models
pinecone.io
vendor site · Pinecone16 answers · 25 citations · 7 models
marktechpost.com
16 answers · 16 citations · 7 models

Pages the answers cite

The ten pages named in the most answers, by full address. A page here is one the models returned with a recommendation, not one the index endorses.

atlan.com/know/top-vector-databases-enterprise-ai
32 answers · 9 models
encore.dev/articles/best-vector-databases
30 answers · 9 models
firecrawl.dev/blog/best-vector-databases
30 answers · 10 models
iternal.ai/insights/best-vector-databases-2026
24 answers · 8 models
sganalytics.com/blog/top-vector-databases-enterprise-ai
21 answers · 9 models
aloa.co/ai/comparisons/vector-database-comparison/best-enterprise-vector-da…
19 answers · 7 models
g2.com/articles/best-vector-databases
18 answers · 8 models
marktechpost.com/2026/05/10/best-vector-databases-in-2026-pricing-scale-lim…
15 answers · 7 models
zenml.io/blog/vector-databases-for-rag
13 answers · 7 models
actian.com/blog/databases/how-to-evaluate-vector-databases-in-2026
10 answers · 8 models

04How they answered

Six framings of the same buying question, each sent to every model in a fresh session with search on. One row per model, so a row shows whether it held its answer under rewording, what it named when cost was the constraint, and what it argued against. Computed from the raw judge labels.
ModelDirect“What is the best vector database for an enterprise B2B company?”Paraphrase“Which vector search database for AI applications would you recommend to a large B2B company with thousands of employees?”Comparative“What are the top enterprise-grade vector databases and how do they differ?”Budget-constrained“What is the best vector database for a large company that needs predictable total cost across thousands of users?”Scale-constrained“We are a 5,000 person company with SSO, SOC 2 and procurement review requirements evaluating a vector database. What should we look at?”Negative“Which vector databases should a large enterprise avoid or be cautious about?”
Claude Haiku 4.5Pinecone
Four alternativesMilvus, Weaviate, Zilliz Cloud, pgvector
PineconeHeld
Three alternativesMilvus, Weaviate, pgvector
Pinecone, Qdrant
Two alternativesMilvus, pgvector
Qdrant, Weaviate
One alternativepgvector
against: Pineconeno first choiceagainst: FAISS, Milvus
GPT-5.4 miniPinecone
Two alternativesMilvus, Weaviate
PineconeHeld
Three alternativesAmazon OpenSearch / Aurora / DynamoDB / Neptune, Google Cloud Vector Search / Agent Retrieval, Qdrant
Pinecone
Four alternativesAzure AI Search, Milvus, MongoDB Atlas Vector Search, Weaviate
against: Elasticsearch / OpenSearch, pgvectorPinecone
Three alternativesMilvus, Qdrant, Weaviate
no first choiceagainst: pgvectornothing named
Gemini 3.5 Flashpgvector
Four alternativesMilvus, Pinecone, Qdrant, Weaviate
PineconeChanged
Four alternativesMilvus, Qdrant, Weaviate, pgvector
pgvector
Three alternativesMilvus, Pinecone, Weaviate
against: Elasticsearch / OpenSearch, RedisQdrant
Two alternativesMilvus / Zilliz Cloud, `pgvector` with `pgvectorscale`
against: Pinecone, Weaviatepgvector
Seven alternativesElasticsearch / OpenSearch, Milvus, MongoDB Atlas Vector Search, Pinecone, Qdrant, Weaviate, Zilliz Cloud
against: Chroma, FAISS, HNSWLIB, LanceDB, Milvus, MongoDB Atlas Vector Search, Pinecone, ScaNN, pgvector
Perplexity SonarPinecone, pgvector
Three alternativesMilvus, Qdrant, Weaviate
PineconeChanged
Four alternativesAzure AI Search, Elasticsearch, Milvus, Weaviate
against: pgvectorPinecone
Three alternativesMilvus / Zilliz Cloud, Qdrant, Weaviate
against: pgvectorQdrant
Two alternativesMixpeek Vector Store, pgvector
no first choiceagainst: Chroma, FAISS, Pinecone, pgvector
Grok 4.1 FastPinecone
Two alternativesWeaviate, Zilliz Cloud
against: Oracle, pgvectorPineconeHeld
Four alternativesMilvus / Zilliz Cloud, Qdrant, Weaviate, pgvector
Pinecone, Zilliz Cloud
Four alternativesMilvus, Qdrant, Weaviate, pgvector
against: Chroma, Elasticsearch, LanceDBQdrant
Two alternativesMilvus, Weaviate
against: Pinecone, Weaviate, Zilliz CloudPinecone
Four alternativesChroma, Qdrant, Weaviate, Zilliz Cloud
against: Milvus, pgvectoragainst: Chroma, Milvus, Pinecone, Weaviate, pgvector
Mistral SmallPinecone, Weaviate
One alternativepgvector
against: MilvusPinecone, WeaviateHeld
Two alternativesAzure AI Search, Milvus
no first choiceMilvus / Zilliz Cloud
One alternativePinecone
against: Weaviateno first choiceagainst: FAISS, Pinecone
DeepSeek V4 FlashPinecone
Four alternativesMilvus / Zilliz Cloud, Qdrant, Weaviate, pgvector
PineconeHeld
Six alternativesAzure AI Search, Milvus / Zilliz Cloud, Qdrant, Vertex AI Vector Search, Weaviate, pgvector
Milvus / Zilliz Cloud, Pinecone, Qdrant, Weaviate
Two alternativesRedis, pgvector
against: Chroma, LanceDBQdrant, Weaviate
One alternativeMilvus
against: Pineconeno first choiceagainst: Chroma, Milvus, Pinecone, Weaviate
Llama 4 Maverickno first choiceno first choiceHeldno first choiceno first choice
Two alternativesQdrant, Weaviate
against: Pinecone, pgvectorPinecone
Four alternativesAzure AI Search, Azure OpenAI Service, Milvus, Weaviate
against: Chroma, Pinecone, pgvector
Qwen 3.7 Flashpgvector
Four alternativesElasticsearch, Milvus, Pinecone, Weaviate
PineconeChanged
Three alternativesMilvus, Weaviate, pgvector
no first choice
Six alternativesElasticsearch, Milvus, Pinecone, Qdrant, Weaviate, Zilliz Cloud
Pinecone
Three alternativesMilvus, Qdrant, pgvector
against: OpenSearch, Pineconeno first choiceagainst: FAISS, Milvus, Numpy, Pinecone, Zilliz Cloud
Kimi K2Pinecone, Zilliz Cloud
Two alternativesWeaviate, pgvector
PineconeChanged
Two alternativesWeaviate, Zilliz Cloud
against: pgvectorPinecone
Three alternativesMilvus / Zilliz Cloud, Qdrant, Weaviate
against: Elasticsearch, pgvectorQdrant
One alternativeMilvus
against: PineconePinecone
Four alternativesMilvus, Qdrant, Weaviate, Zilliz Cloud
against: pgvectoragainst: Annoy, Chroma, FAISS, Pinecone, Qdrant, Weaviate
GLM 4.7 FlashXPinecone
Four alternativesMilvus / Zilliz Cloud, Qdrant, Weaviate, pgvector
QdrantChanged
Three alternativesMilvus / Zilliz Cloud, Pinecone, Weaviate
against: pgvectorPinecone
Eight alternativesElasticsearch, Milvus / Zilliz Cloud, MongoDB Atlas Vector Search, Qdrant, Redis, Vespa, Weaviate, pgvector
against: ChromaZilliz Cloud
Two alternativesQdrant, pgvector
against: Pineconeno first choiceagainst: Chroma, Elasticsearch, MongoDB Atlas Vector Search, Redis, pgvector
MiniMax M2.5pgvector
Two alternativesMilvus, Weaviate
against: PineconePineconeChanged
Three alternativesAzure AI Search, Weaviate, pgvector
Pinecone
Seven alternativesElasticsearch, Milvus / Zilliz Cloud, Qdrant, Redis, Vespa, Weaviate, pgvector
Qdrant
Two alternativesPinecone, Zilliz Cloud
against: Upstash Vectorno first choiceagainst: Milvus, Pinecone, Weaviate, pgvector
Bold is the first choiceAlternatives are counted; the count opens them.What the answer argued against

05The record

One row per call: the version string exactly as returned, whether the model searched, sources cited, and latency. Full answer text is in the free responses file. Download the record
Seventy-two rows: every prompt, every model, every answer.
PromptModelVersion stringTime (UTC)SearchedSourcesLatency
Direct recommendationClaude Haiku 4.5claude-haiku-4-5-202510012026-09-17 10:38yes1611 s
Direct recommendationGPT-5.4 minigpt-5.4-mini-2026-03-172026-09-17 11:44yes57 s
Direct recommendationGemini 3.5 Flashgemini-3.5-flash2026-09-17 11:00yes1126 s
Direct recommendationPerplexity Sonarsonar2026-09-17 09:27yes206 s
Direct recommendationGrok 4.1 Fastspacexai/grok-4.1-fast-non-reasoning via vertex2026-09-17 13:22yes3821 s
Direct recommendationMistral Smallmistral/mistral-small via mistral2026-09-17 14:02yes55 s
Direct recommendationDeepSeek V4 Flashdeepseek/deepseek-v4-flash via fireworks2026-09-17 12:49yes1015 s
Direct recommendationLlama 4 Maverickmeta/llama-4-maverick via bedrock2026-09-17 13:32yes53 s
Direct recommendationQwen 3.7 Flashalibaba/qwen3.7-flash via alibaba2026-09-17 10:16no043 s
Direct recommendationKimi K2moonshotai/kimi-k2 via novita2026-09-17 13:35yes1533 s
Direct recommendationGLM 4.7 FlashXzai/glm-4.7-flashx via zai2026-09-17 10:09yes1545 s
Direct recommendationMiniMax M2.5minimax/minimax-m2.5 via minimax2026-09-17 12:13yes922 s
ParaphraseClaude Haiku 4.5claude-haiku-4-5-202510012026-09-17 10:41yes98 s
ParaphraseGPT-5.4 minigpt-5.4-mini-2026-03-172026-09-17 09:37yes45 s
ParaphraseGemini 3.5 Flashgemini-3.5-flash2026-09-17 13:56yes822 s
ParaphrasePerplexity Sonarsonar2026-09-17 10:45yes205 s
ParaphraseGrok 4.1 Fastspacexai/grok-4.1-fast-non-reasoning via vertex2026-09-17 11:26yes3716 s
ParaphraseMistral Smallmistral/mistral-small via mistral2026-09-17 09:58yes56 s
ParaphraseDeepSeek V4 Flashdeepseek/deepseek-v4-flash via fireworks2026-09-17 13:34yes1923 s
ParaphraseLlama 4 Maverickmeta/llama-4-maverick via bedrock2026-09-17 11:35yes53 s
ParaphraseQwen 3.7 Flashalibaba/qwen3.7-flash via alibaba2026-09-17 12:21no040 s
ParaphraseKimi K2moonshotai/kimi-k2 via novita2026-09-17 12:16yes1984 s
ParaphraseGLM 4.7 FlashXzai/glm-4.7-flashx via zai2026-09-17 13:10yes1524 s
ParaphraseMiniMax M2.5minimax/minimax-m2.5 via minimax2026-09-17 14:02yes941 s
ComparativeClaude Haiku 4.5claude-haiku-4-5-202510012026-09-17 10:17yes99 s
ComparativeGPT-5.4 minigpt-5.4-mini-2026-03-172026-09-17 11:59yes712 s
ComparativeGemini 3.5 Flashgemini-3.5-flash2026-09-17 12:32yes2331 s
ComparativePerplexity Sonarsonar2026-09-17 12:15yes2012 s
ComparativeGrok 4.1 Fastspacexai/grok-4.1-fast-non-reasoning via vertex2026-09-17 12:27yes2114 s
ComparativeMistral Smallmistral/mistral-small via mistral2026-09-17 11:59yes58 s
ComparativeDeepSeek V4 Flashdeepseek/deepseek-v4-flash via fireworks2026-09-17 09:54yes1823 s
ComparativeLlama 4 Maverickmeta/llama-4-maverick via bedrock2026-09-17 11:40yes53 s
ComparativeQwen 3.7 Flashalibaba/qwen3.7-flash via alibaba2026-09-17 13:05no040 s
ComparativeKimi K2moonshotai/kimi-k2 via novita2026-09-17 13:32yes1536 s
ComparativeGLM 4.7 FlashXzai/glm-4.7-flashx via zai2026-09-17 11:00yes1382 s
ComparativeMiniMax M2.5minimax/minimax-m2.5 via minimax2026-09-17 10:38yes519 s
Budget constrainedClaude Haiku 4.5claude-haiku-4-5-202510012026-09-17 12:00yes1411 s
Budget constrainedGPT-5.4 minigpt-5.4-mini-2026-03-172026-09-17 09:37yes35 s
Budget constrainedGemini 3.5 Flashgemini-3.5-flash2026-09-17 12:11yes1326 s
Budget constrainedPerplexity Sonarsonar2026-09-17 13:41yes179 s
Budget constrainedGrok 4.1 Fastspacexai/grok-4.1-fast-non-reasoning via vertex2026-09-17 14:02yes4620 s
Budget constrainedMistral Smallmistral/mistral-small via mistral2026-09-17 10:49yes58 s
Budget constrainedDeepSeek V4 Flashdeepseek/deepseek-v4-flash via fireworks2026-09-17 10:08yes1619 s
Budget constrainedLlama 4 Maverickmeta/llama-4-maverick via bedrock2026-09-17 11:49yes54 s
Budget constrainedQwen 3.7 Flashalibaba/qwen3.7-flash via alibaba2026-09-17 11:37yes535 s
Budget constrainedKimi K2moonshotai/kimi-k2 via novita2026-09-17 11:03yes1430 s
Budget constrainedGLM 4.7 FlashXzai/glm-4.7-flashx via zai2026-09-17 11:50yes2777 s
Budget constrainedMiniMax M2.5minimax/minimax-m2.5 via minimax2026-09-17 12:07yes1567 s
Scale constrainedClaude Haiku 4.5claude-haiku-4-5-202510012026-09-17 13:16yes2711 s
Scale constrainedGPT-5.4 minigpt-5.4-mini-2026-03-172026-09-17 14:12yes514 s
Scale constrainedGemini 3.5 Flashgemini-3.5-flash2026-09-17 10:38no022 s
Scale constrainedPerplexity Sonarsonar2026-09-17 09:39yes2017 s
Scale constrainedGrok 4.1 Fastspacexai/grok-4.1-fast-non-reasoning via vertex2026-09-17 10:26yes1812 s
Scale constrainedMistral Smallmistral/mistral-small via mistral2026-09-17 14:05no07 s
Scale constrainedDeepSeek V4 Flashdeepseek/deepseek-v4-flash via fireworks2026-09-17 12:06yes2639 s
Scale constrainedLlama 4 Maverickmeta/llama-4-maverick via bedrock2026-09-17 13:10yes55 s
Scale constrainedQwen 3.7 Flashalibaba/qwen3.7-flash via alibaba2026-09-17 12:21yes537 s
Scale constrainedKimi K2moonshotai/kimi-k2 via novita2026-09-17 13:41yes1885 s
Scale constrainedGLM 4.7 FlashXzai/glm-4.7-flashx via zai2026-09-17 10:44yes1432 s
Scale constrainedMiniMax M2.5minimax/minimax-m2.5 via minimax2026-09-17 13:31yes532 s
Negative framingClaude Haiku 4.5claude-haiku-4-5-202510012026-09-17 11:10yes98 s
Negative framingGPT-5.4 minigpt-5.4-mini-2026-03-172026-09-17 10:36yes69 s
Negative framingGemini 3.5 Flashgemini-3.5-flash2026-09-17 11:05yes2230 s
Negative framingPerplexity Sonarsonar2026-09-17 11:32yes208 s
Negative framingGrok 4.1 Fastspacexai/grok-4.1-fast-non-reasoning via vertex2026-09-17 10:48yes2019 s
Negative framingMistral Smallmistral/mistral-small via mistral2026-09-17 13:15yes58 s
Negative framingDeepSeek V4 Flashdeepseek/deepseek-v4-flash via fireworks2026-09-17 12:46yes1630 s
Negative framingLlama 4 Maverickmeta/llama-4-maverick via bedrock2026-09-17 10:16yes54 s
Negative framingQwen 3.7 Flashalibaba/qwen3.7-flash via alibaba2026-09-17 10:28yes547 s
Negative framingKimi K2moonshotai/kimi-k2 via novita2026-09-17 14:01yes2944 s
Negative framingGLM 4.7 FlashXzai/glm-4.7-flashx via zai2026-09-17 11:56yes1472 s
Negative framingMiniMax M2.5minimax/minimax-m2.5 via minimax2026-09-17 11:52yes1438 s

Normalization in this category

Every judgment call made between the raw labels and the numbers above, listed so it is visible and reversible.

Category-scoped readings
MongoDB read as MongoDB Atlas Vector Search
MongoDB Atlas read as MongoDB Atlas Vector Search
PostgreSQL (+ pgvector) read as pgvector
PostgreSQL (via pgvector) read as pgvector
Unresolved, counted raw
Actian VectorAI DB
Amazon OpenSearch / Aurora / DynamoDB / Neptune
Databricks Vector Search
Google Cloud Vector Search / Agent Retrieval
HNSWLIB
Mixpeek Vector Store (MVS)
Numpy
ScaNN
TiDB Cloud Serverless
Upstash Vector
`pgvector` with `pgvectorscale`
Discontinued, still offered
No shut-down product was recommended here.

Also in data platform

The other categories of this function, with the leader in each.
← StreamingOrchestration →