{"slug":"pgvector","name":"pgvector","domain":"github.com","verdict":"As of 2026-07-15, ChatGPT, Claude, Gemini collectively rank pgvector #2 of 7 for vector database for production ai apps (one of 5 leaderboards it appears on). Source: https://modelsagree.com/product/pgvector (modelsagree.com, CC BY 4.0).","best_rank":2,"categories":5,"brief":{"category":"best-vector-database","title":"Best vector database for production AI apps","rank":2,"of":7,"top":"Qdrant","day":"2026-07-17","why":[{"t":"No new system to operate","m":["Claude","Gemini","ChatGPT"],"q":"no new system to operate"},{"t":"Transactions, joins, and SQL filtering","m":["Claude","Gemini","ChatGPT"],"q":"PostgreSQL transactions, joins, SQL filtering"},{"t":"HNSW with mature operational tooling","m":["Claude","Gemini","ChatGPT"],"q":"mature operational tooling, exact search, HNSW, IVFFlat, iterative scans"}],"gap":[{"t":"Fast filtered vector search","m":["ChatGPT","Claude","Gemini"],"q":"fast filtered vector search"},{"t":"Built-in quantization and hybrid support","m":["ChatGPT","Claude","Gemini"],"q":"built-in quantization and hybrid/sparse support cut memory cost sharply"},{"t":"Flexible self-hosted or managed deployment","m":["ChatGPT","Claude"],"q":"flexible open-source, self-hosted, or managed deployment"}],"fix":[{"t":"Not for extremely large workloads","m":["ChatGPT","Claude","Gemini"],"q":"It is not the best default for extremely large, vector-heavy workloads"},{"t":"Filtered high-QPS performance falls behind","m":["Claude"],"q":"Not for very large or high-QPS filtered workloads"},{"t":"Vector operations deplete SQL resources","m":["Gemini"],"q":"vector operations can deplete CPU and memory resources required for transactional SQL queries"}]},"entries":[{"slug":"best-vector-database","title":"Best vector database for production AI apps","rank":2,"of":7,"score":12,"appearances":3,"modelRanks":{"ChatGPT":4,"Claude":1,"Gemini":1},"reason":"For the typical practitioner it wins on total cost of adoption — vectors live beside the app's relational data with real transactions, joins, and backups, no new system to operate; pgvectorscale/HNSW handles the tens-of-millions-of-vectors range most production RAG apps actually reach, and every managed Postgres (RDS, Supabase, Neon) ships it; ranked #1 on the assumption the typical app is well under ~100M vectors","reasons":[{"model":"Claude","reason":"For the typical practitioner it wins on total cost of adoption — vectors live beside the app's relational data with real transactions, joins, and backups, no new system to operate; pgvectorscale/HNSW handles the tens-of-millions-of-vectors range most production RAG apps actually reach, and every managed Postgres (RDS, Supabase, Neon) ships it; ranked #1 on the assumption the typical app is well under ~100M vectors"},{"model":"Gemini","reason":"Minimizes operational overhead for the typical practitioner by extending existing PostgreSQL setups. It supports ACID compliance, relational joins, HNSW indexing, and avoids introducing new infrastructure. Ranked first under the assumption that most production apps already use a relational database and prefer to minimize complex architectural footprints."},{"model":"ChatGPT","reason":"Best value when vectors belong beside relational application data: PostgreSQL transactions, joins, SQL filtering, mature operational tooling, exact search, HNSW, IVFFlat, iterative scans, and no separate database to synchronize."}],"fixes":[{"model":"ChatGPT","fix":"It is not the best default for extremely large, vector-heavy workloads requiring effortless horizontal sharding and independently scalable retrieval throughput."},{"model":"Claude","fix":"Not for very large or high-QPS filtered workloads — past ~100M vectors or heavy metadata-filtered ANN, index build times, memory pressure, and recall/latency tuning fall behind dedicated engines"},{"model":"Gemini","fix":"Not built for multi-billion vector scales or high-concurrency, ultra-low-latency workloads where vector operations can deplete CPU and memory resources required for transactional SQL queries."}],"updated":"2026-07-15","rank_history":{"days":["2026-06-29","2026-07-08","2026-07-09","2026-07-10","2026-07-12","2026-07-13","2026-07-14","2026-07-15"],"ranks":[5,4,4,3,3,1,2,3]},"reasoning_shift":[{"model":"Gemini","from":"2026-07-14","to":"2026-07-15","added":[{"t":"HNSW indexing","q":"HNSW indexing"},{"t":"High-concurrency latency limits","q":"high-concurrency, ultra-low-latency workloads"}],"dropped":[{"t":"Metadata filtering","q":"metadata filtering"},{"t":"Tens of millions scaling","q":"scaling to tens of millions of vectors"},{"t":"Dedicated engine optimizations","q":"query-level optimizations and distributed clustering of dedicated vector engines"}]},{"model":"ChatGPT","from":"2026-07-14","to":"2026-07-15","added":[{"t":"Exact search","q":"exact search"},{"t":"Iterative scans","q":"iterative scans"},{"t":"Effortless horizontal sharding","q":"effortless horizontal sharding"}],"dropped":[{"t":"Specialized retrieval features","q":"specialized retrieval features"}]}],"api":"https://modelsagree.com/api/v1/best/best-vector-database.json"},{"slug":"best-vector-database-for-production-rag","title":"Best Vector database for production RAG","rank":3,"of":6,"score":11,"appearances":4,"modelRanks":{"ChatGPT":4,"Claude":2,"Gemini":2,"Grok":5},"reason":"For teams already on Postgres, the strongest real-world value — vectors live next to relational data, so joins, transactions, and exact metadata filtering are free; pgvectorscale's StreamingDiskANN and quantization closed most of the performance gap; zero new infrastructure to operate, works on RDS/Supabase/Neon. Near-tie with Qdrant — rank flips if you already run Postgres.","reasons":[{"model":"Claude","reason":"For teams already on Postgres, the strongest real-world value — vectors live next to relational data, so joins, transactions, and exact metadata filtering are free; pgvectorscale's StreamingDiskANN and quantization closed most of the performance gap; zero new infrastructure to operate, works on RDS/Supabase/Neon. Near-tie with Qdrant — rank flips if you already run Postgres."},{"model":"Gemini","reason":"Delivers the highest practical ROI for most production RAG pipelines by running vector operations directly inside PostgreSQL, eliminating dual-write sync errors and complex infrastructure overhead; near-tie with Qdrant for overall top rank."},{"model":"ChatGPT","reason":"Best value when application data already lives in PostgreSQL; preserves transactions, joins, SQL filtering, backups, and existing operational tooling while providing exact search plus HNSW and IVFFlat indexes"},{"model":"Grok","reason":"Seamless integration with existing Postgres (no new DB), ACID + SQL joins for metadata/docs, sufficient for <50-100M vectors in many production RAG setups with low ops overhead."}],"fixes":[{"model":"ChatGPT","fix":"Not the best fit for independently scaling very large, high-throughput vector workloads where vector-native distributed systems are easier to tune"},{"model":"Claude","fix":"Not for high-QPS, billion-vector dedicated workloads — index build times, memory pressure on shared instances, and recall/latency ceilings show up well before purpose-built engines'."},{"model":"Gemini","fix":"Not for multi-billion vector scales or hyper-dense write-heavy workloads that exceed traditional relational database index capacities."}],"updated":"2026-07-19","api":"https://modelsagree.com/api/v1/best/best-vector-database-for-production-rag.json"},{"slug":"best-vector-search-services-for-multi-tenant-saas","title":"Best vector search services for multi-tenant SaaS","rank":5,"of":6,"score":4,"appearances":2,"modelRanks":{"Claude":4,"Gemini":4},"reason":"If tenant data already lives in Postgres (as it does for most SaaS), row-level security gives airtight per-tenant isolation, vectors stay transactional with the rest of the tenant's data, and managed options (Supabase, Neon, RDS) make it near-zero extra infrastructure — the right default below ~10M vectors per instance.","reasons":[{"model":"Claude","reason":"If tenant data already lives in Postgres (as it does for most SaaS), row-level security gives airtight per-tenant isolation, vectors stay transactional with the rest of the tenant's data, and managed options (Supabase, Neon, RDS) make it near-zero extra infrastructure — the right default below ~10M vectors per instance."},{"model":"Gemini","reason":"Leverages existing PostgreSQL databases, allowing developers to enforce tenant isolation via native SQL features like Row-Level Security (RLS) or schema-per-tenant isolation, completely removing the operational and synchronization overhead of managing a separate vector database."}],"fixes":[{"model":"Claude","fix":"HNSW index build times and memory pressure become painful past tens of millions of vectors, and recall/latency under heavy per-tenant filtering trails dedicated engines — not for large-scale or high-QPS vector workloads."},{"model":"Gemini","fix":"Heavy HNSW indexing and high-concurrency query workloads consume massive CPU/RAM, which can easily degrade or crash the main transactional relational database."}],"updated":"2026-07-16","api":"https://modelsagree.com/api/v1/best/best-vector-search-services-for-multi-tenant-saas.json"},{"slug":"best-hybrid-search-engines-for-on-premises-rag","title":"Best hybrid search engines for on-premises RAG","rank":6,"of":7,"score":2,"appearances":1,"modelRanks":{"Gemini":4},"reason":"Enables hybrid search directly inside an existing relational engine by combining native full-text search (tsvector/BM25) with pgvector dense embeddings. Ranked fourth assuming that eliminating infrastructure sprawl yields the highest net practical value for small-to-midscale on-prem environments.","reasons":[{"model":"Gemini","reason":"Enables hybrid search directly inside an existing relational engine by combining native full-text search (tsvector/BM25) with pgvector dense embeddings. Ranked fourth assuming that eliminating infrastructure sprawl yields the highest net practical value for small-to-midscale on-prem environments."}],"fixes":[{"model":"Gemini","fix":"Experiences performance bottlenecks in query latency and index build speed when scaling to tens of millions of high-dimensional vectors under high write concurrency."}],"updated":"2026-08-07","api":"https://modelsagree.com/api/v1/best/best-hybrid-search-engines-for-on-premises-rag.json"},{"slug":"best-long-term-memory-stores-for-ai-agents","title":"Best long-term memory stores for AI agents","rank":9,"of":10,"score":1,"appearances":1,"modelRanks":{"Claude":5},"reason":"The build-it-yourself baseline that a large share of production agents actually run on — memory rows with embeddings, metadata, and recency/importance scoring in the database you already operate; zero new vendors, real transactions, trivially auditable, and mature ops tooling. Earns the spot on total-cost-of-ownership for teams with existing Postgres competence.","reasons":[{"model":"Claude","reason":"The build-it-yourself baseline that a large share of production agents actually run on — memory rows with embeddings, metadata, and recency/importance scoring in the database you already operate; zero new vendors, real transactions, trivially auditable, and mature ops tooling. Earns the spot on total-cost-of-ownership for teams with existing Postgres competence."}],"fixes":[{"model":"Claude","fix":"You get storage, not memory management — extraction, consolidation, forgetting, and contradiction handling are all on you, which is exactly the hard part the top picks solve."}],"updated":"2026-07-17","api":"https://modelsagree.com/api/v1/best/best-long-term-memory-stores-for-ai-agents.json"}],"page":"https://modelsagree.com/product/pgvector","check":"https://modelsagree.com/check?q=pgvector","updated":"2026-08-10T18:18:45.051Z","attribution":"modelsagree.com, CC BY 4.0"}