Open-source SuperEMeans combines SuperKMeans and e-means for 3x faster clustering
Developers at the team behind SuperKMeans have released SuperEMeans, a method that merges their dimension-pruning clustering technique with the recently published e-means algorithm. The combined approach clusters high-dimensional vector embeddings roughly three times faster than existing methods without sacrificing clustering quality, and the code is available on GitHub.
GoKawiil's interpretation of the reporting above, not reported fact.
Faster clustering directly speeds up the construction of large-scale vector indexes like IVF and ScANN, which underpin similarity search systems used in retrieval and recommendation pipelines. Because SuperKMeans has already been adopted by projects such as ParadeDB, Zilliz, and FAISS, a compatible speed boost via e-means integration could see similarly quick uptake across the vector search ecosystem. The authors frame the two techniques as complementary rather than competing, suggesting further combined optimizations may follow.
- SuperEMeans merges SuperKMeans' dimension-pruning with e-means' reduced-epoch approach for a 3x speedup.
- The method targets clustering of high-dimensional vector embeddings used in large-scale vector search indexes.
- Code is open-sourced on GitHub, building on SuperKMeans' prior adoption by industry tools like FAISS and Zilliz.
Source: lkuffo.com — Leonardo Kuffo, 2026-11-18
Published there as: “SuperEMeans: Making e-means 3× faster”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.