Google Research open-sources ScaNN

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Open-sourced ScaNN — applies anisotropic vector quantization; on ann-benchmarks, about 2x the QPS of 11 other libraries at the same accuracy

Why it matters

Redesigned the quantization loss function around inner-product estimation error, pushing the quality-speed frontier of MIPS forward by a generation

Source — Primary sources first; secondary reports are labelled “Press”

Officialresearch.googlehttps://research.google/blog/announcing-scann-efficient-vector-similarity-search/

Added in Week 35, 2026 · See that week

On the map — ⑰ Embeddings line

  1. Previous stopDPR
  2. This recordScaNN
  3. Next stopMTEB
See it on the map

Nearby on the same axis

Karpukhin et al. present Dense Passage RetrievalPaperShen et al. present Towards Backward-Compatible Representation LearningPaperPinecone launches managed vector database beta, announces $10 million seedOfficialMicrosoft Research India and CMU present DiskANNPaper