February Release: Performance at Scale, Predictability, and Control

This post details significant performance improvements in Pinecone's vector search, particularly for large indexes. Engineering efforts included rewriting core parts of the engine in Rust, optimizing I/O operations, implementing dynamic caching, and re-configuring storage formats. Benchmark tests show substantial latency reductions for both performance-optimized (p1) and storage-optimized (s1) pods as index size increases. The release also introduces a new usage estimator tool, reduces variance between p50 and p95 search latencies, and makes data ingestion faster and more reliable. Additionally, it offers greater control over projects and environments with new deployment options across GCP and AWS for Standard plan users, and a redesigned management console.