# Vector databases: what AI models recommend, September 2026

Gane Index, September 2026 Edition. Asked as "vector database" and as "vector search database for AI applications", six framings each, to twelve models with search on, on behalf of a mid-market B2B company. Page: https://www.gane.ai/index/it/it-data/vector-databases/2026-09

This is the permanent record of the September 2026 Edition, published September 17, 2026 and superseded since. The current standing: https://www.gane.ai/index/it/it-data/vector-databases/index.md

**Standing:** Qdrant leads with 37% of first choices; verdict contested. 51 first choices across the direct, paraphrase, budget and scale prompts.

## First-choice share

| Rank | Product | Share | Negative rate | Labels |
|---|---|---|---|---|
| 1 | [Qdrant](https://www.gane.ai/index/it/vendors/qdrant/index.md) | 37% | 5% | 63 |
| 2 | [pgvector](https://www.gane.ai/index/it/vendors/pgvector/index.md) | 35% | 14% | 51 |
| 3 | [Pinecone](https://www.gane.ai/index/it/vendors/pinecone/index.md) | 14% | 28% | 58 |
| 4 | [Weaviate](https://www.gane.ai/index/it/vendors/weaviate/index.md) | 6% | 15% | 61 |
| 5 | [Chroma](https://www.gane.ai/index/it/vendors/chroma/index.md) | 4% | 27% | 30 |
| 6 | [Milvus](https://www.gane.ai/index/it/vendors/milvus/index.md) | 2% | 37% | 38 |
| 11 | [FAISS](https://www.gane.ai/index/it/vendors/faiss/index.md) | 0% | 18% | 11 |
| 12 | [Milvus / Zilliz Cloud](https://www.gane.ai/index/it/vendors/milvus-zilliz-cloud/index.md) | 0% | 36% | 14 |

Rank is by share of first choices among all 24 products named; products with the same share are tied, and the order among them carries no meaning. The table lists those with at least 10 labels.

## Each model's first choice on the direct prompt

- Claude Haiku 4.5: Pinecone, Qdrant; alternatives Weaviate, pgvector
- GPT-5.4 mini: Qdrant, pgvector; alternatives Milvus / Zilliz Cloud, Pinecone, Weaviate
- Gemini 3.5 Flash: pgvector; alternatives Pinecone, Qdrant, Weaviate
- Perplexity Sonar: Qdrant, pgvector; alternatives Milvus, Pinecone, Weaviate
- Grok 4.20: pgvector; alternatives Pinecone, Qdrant
- Mistral Small: pgvector; alternatives Pinecone, Qdrant, Weaviate
- DeepSeek V4 Flash: Qdrant, Weaviate; alternatives pgvector
- Llama 4 Maverick: pgvector
- Qwen 3.7 Flash: pgvector; alternatives Pinecone, Qdrant, Weaviate
- Kimi K2: Pinecone, pgvector; alternatives Qdrant, Weaviate
- GLM 4.7 FlashX: pgvector; alternatives Pinecone, Qdrant, Weaviate
- MiniMax M2.5: pgvector; alternatives Pinecone, Qdrant, Weaviate

## Sources the answers cite

68 of 72 answers came back with a source list, from 12 of 12 models. Sites named in the most answers:

- firecrawl.dev: 47 answers, 47 citations
- dev.to: 28 answers, 35 citations
- encore.dev: 28 answers, 28 citations
- redis.io: 24 answers, 25 citations
- medium.com: 18 answers, 24 citations
- marktechpost.com: 18 answers, 19 citations
- iternal.ai: 17 answers, 17 citations
- zenml.io: 17 answers, 17 citations

Pages named in the most answers:

- https://firecrawl.dev/blog/best-vector-databases (47 answers)
- https://encore.dev/articles/best-vector-databases (28 answers)
- https://iternal.ai/insights/best-vector-databases-2026 (17 answers)
- https://redis.io/blog/best-open-source-vector-databases-comparison (17 answers)
- https://zenml.io/blog/vector-databases-for-rag (17 answers)
- https://g2.com/articles/best-vector-databases (14 answers)
- https://marktechpost.com/2026/05/10/best-vector-databases-in-2026-pricing-scale-limits-and-architecture-tradeoffs-across-nine-leading-systems (14 answers)
- https://mastra.ai/articles/best-vector-database-providers (13 answers)
- https://actian.com/blog/databases/how-to-evaluate-vector-databases-in-2026 (12 answers)
- https://liveblocks.io/blog/whats-the-best-vector-database-for-building-ai-products (12 answers)

## Warned against

- Pinecone: 16 of 58 labels negative. "Specific Targets to Watch Out For: Pinecone (purely managed serverless)... you cannot take your data elsewhere without significant re-architecture." (Qwen 3.7 Flash, negative prompt)
- Milvus: 14 of 38 labels negative. "What to Avoid (For Now)... Milvus / Vespa... generally overkill for mid-market requirements." (Gemini 3.5 Flash, direct prompt)
- Weaviate: 9 of 61 labels negative. "Avoid Initially: Pinecone/Weaviate Cloud (managed convenience costs $25-70/mo min after free tiers" (Grok 4.20, budget prompt)
- pgvector: 7 of 51 labels negative. "When you need very high-scale, pure vector search | It is often recommended for existing Postgres shops, but can be a weaker fit for large-scale semantic search workloads." (Perplexity Sonar, negative prompt)

## Record

- Method: https://www.gane.ai/index/it/methodology
- License: CC BY 4.0. Cite as Gane Index, September 2026 Edition, www.gane.ai/index/it.