Learn about our story to get to the automated performance testing.

Automating performance tests

Use Adaptive Retrieval to improve query performance with OpenAI's new embedding models

Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval

pgvector 0.6.0 brings a significant improvement: parallel index builds for HNSW. Building an HNSW index is now up to 30x faster for unlogged tables.

pgvector 0.6.0: 30x faster with parallel index builds

Direct performance comparison between pgvector and Pinecone.

pgvector vs Pinecone: cost and performance

Increase performance in pgvector using HNSW indexes

pgvector v0.5.0: Faster semantic search with HNSW indexes

Increase performance in pgvector by using embedding vectors with fewer dimensions

pgvector: Fewer dimensions are better

There's been a lot of talk about pgvector performance lately, so we took some datasets and pushed pgvector to the limits to find out its strengths and limitations.