PII Arena
OPF detector

opf-ru-v2+homoglyph PII detection benchmark

Measured on 3 of 41 retained datasets under the same frozen scoring and masking protocol.

51.03%fully hidden annotations
66.48%detected / overlapped
2.76%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Passwords, keys & tokens76.24%292 / 383
Phone numbers & email65.98%514 / 779
People's names57.73%859 / 1,488

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Network identifiers26.04%94 / 361
Addresses & locations26.94%382 / 1,418
Bank accounts & cards52.15%194 / 372

Language coverage

The overall leader can differ from the best choice for one language slice.

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English170.14%92.71%3.57%
Russian250.26%65.43%0.93%

Configuration, source and version

Family
OPF
Revision
7975e895ed59a7c58e9ab2dbb7d739fd246086b4
Eligible datasets
3 / 41
Character F1
0.564
Untouched diagnostic
33.52% · 2,509 / 7,486
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
Not measured
GPU seconds / 10k chars
Not measured

Results by dataset

Complete masking is primary; detection shows whether any character overlapped.

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
corrupt-redmadrobot5,53144.71%58.90%0.86%
corrupt-hivetrace1,66768.69%87.10%1.44%
corrupt-secrets-issues28870.14%92.71%3.57%