Benchmarks
This page contains the same benchmark data as the top-level Benchmarks page, formatted for quick reference within the guide.
Summary
| Operation | p50 | p95 |
|---|---|---|
| Vector search (FAISS) | 238us | 545us |
| FTS search (BM25) | 547us | 800us |
| Hybrid search (RRF) | 1.31ms | 1.37ms |
| Graph traversal (2 hops) | 87us | 374us |
| Dedup check (MinHash) | 1.25ms | -- |
vs Competitors
| System | Vector search (10K) |
|---|---|
| Ariadne (FAISS) | 0.24ms |
| ChromaDB | 2.39ms |
| sqlite-vec | 0.99ms |
- 10x faster than ChromaDB
- 4.2x faster than sqlite-vec
Scaling
| Dataset | Vector | FTS | Hybrid |
|---|---|---|---|
| 100 | 0.05ms | 0.3ms | 0.6ms |
| 1,000 | 0.24ms | 0.55ms | 1.31ms |
| 10,000* | 0.30ms | 0.55ms | 1.2ms |
| 100,000* | 0.8ms | 1.8ms | 4.5ms |
| 1,000,000* | 3.2ms | 8.0ms | 18ms |
*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