Skip to content

Latest commit

 

History

History
105 lines (99 loc) · 3.07 KB

File metadata and controls

105 lines (99 loc) · 3.07 KB

Codegraph Embedding Benchmarks

Self-measured on every release using auto-generated queries from symbol names. Each symbol's name is split into words (e.g. buildGraph"build graph") and used as the search query. Hit@N = expected symbol found in top N results.

Version Model Symbols Hit@1 Hit@3 Hit@5 Misses Embed Time
3.9.6 minilm 1500 67.4% 87.2% 92.4% 49 125.9s
3.9.6 jina-small 1500 75.8% 93.6% 96.5% 23 253.8s
3.9.6 jina-base 1500 72.1% 90.3% 94.9% 30 1327.4s
3.9.6 nomic 1500 80.1% 95.5% 97.9% 15 1331.2s
3.9.6 nomic-v1.5 1500 78.4% 94.9% 97.1% 18 1325.3s

Latest results

Version: 3.9.6 | Strategy: structured | Symbols: 1500 | Date: 2026-04-30

Model Dim Context Hit@1 Hit@3 Hit@5 Hit@10 Misses Embed Search
minilm 384 256 67.4% 87.2% 92.4% 96.7% 49 125.9s 130.5s
jina-small 512 8192 75.8% 93.6% 96.5% 98.5% 23 253.8s 159.2s
jina-base 768 8192 72.1% 90.3% 94.9% 98.0% 30 1327.4s 206.2s
nomic 768 8192 80.1% 95.5% 97.9% 99.0% 15 1331.2s 206.0s
nomic-v1.5 768 8192 78.4% 94.9% 97.1% 98.8% 18 1325.3s 207.0s