VIBE: Vector Index Benchmark for Embeddings

Authors

Affiliations

Elias Jääsaari

University of Helsinki

Ville Hyvönen

University of Helsinki

Matteo Ceccarello

University of Padova

Teemu Roos

University of Helsinki

Martin Aumüller

IT University of Copenhagen

This website hosts the results of VIBE, the Vector Index Benchmark for Embeddings. Use its interactive plots to explore the benchmark results.

Explore the results

Overview Compare every algorithm across all datasets at a glance. Datasets Explore the datasets, embeddings, dimensions, and distance measures. Tradeoffs Examine the tradeoff between recall and query throughput. Algorithm focus Inspect every configuration tested for each algorithm. Robustness Investigate how algorithms respond to query difficulty.

Run VIBE

The benchmark code and instructions for running it are available on GitHub.

Extend VIBE

Interested in contributing? Follow the guide for adding a new method.

Citation

If you find our work useful, please cite:

E. Jääsaari, V. Hyvönen, M. Ceccarello, T. Roos, M. Aumüller. VIBE: Vector Index Benchmark for Embeddings. Journal of Data-centric Machine Learning Research, 2026.