Skip to content

Benchmarks

This page contains the same benchmark data as the top-level Benchmarks page, formatted for quick reference within the guide.

Summary

Operationp50p95
Vector search (FAISS)238us545us
FTS search (BM25)547us800us
Hybrid search (RRF)1.31ms1.37ms
Graph traversal (2 hops)87us374us
Dedup check (MinHash)1.25ms--

vs Competitors

SystemVector search (10K)
Ariadne (FAISS)0.24ms
ChromaDB2.39ms
sqlite-vec0.99ms
  • 10x faster than ChromaDB
  • 4.2x faster than sqlite-vec

Scaling

DatasetVectorFTSHybrid
1000.05ms0.3ms0.6ms
1,0000.24ms0.55ms1.31ms
10,000*0.30ms0.55ms1.2ms
100,000*0.8ms1.8ms4.5ms
1,000,000*3.2ms8.0ms18ms

*Estimated

Hardware

  • 4-core VPS, 8GB RAM, Ubuntu 24.04
  • 72GB SSD
  • Python 3.11.15
  • ONNX all-MiniLM-L6-v2 (384-dim)

Full benchmark code: benchmarks/run.py

Released under the MIT License.