PII Arena
GLINER2 detector

gliner25-fastino-ru PII detection benchmark

Measured on 19 of 41 retained datasets under the same frozen scoring and masking protocol.

64.57%fully hidden annotations
73.29%detected / overlapped
3.37%extra masking
1,301CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Phone numbers & email88.63%12,743 / 14,377
People's names79.16%26,862 / 33,934
Network identifiers61.88%14,289 / 23,092

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Customer & employee IDs25.52%1,030 / 4,036
Dates & times26.89%709 / 2,637
Other sensitive attributes29.20%325 / 1,113

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
Russian1964.57%73.29%3.37%

Configuration, source and version

Family
GLINER2
Revision
aaecfe45db1d828c963717054ccb868e8ad1f1d5
Eligible datasets
19 / 41
Character F1
0.792
Untouched diagnostic
26.71% · 36,443 / 136,419
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
7.686
GPU seconds / 10k chars
Not measured

Russian labels

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
rubai-ru3,4587.87%97.83%
alexen21,26127.12%99.37%9.34%
corrupt-redmadrobot5,53147.46%50.55%0.50%
synth-secrets-ru58151.98%54.39%14.51%
scanpatch8,70852.85%56.45%2.15%
nym-ru9,49755.14%57.39%
redmadrobot5,51657.92%60.26%0.45%
multiconer-ru1,20859.19%78.56%12.70%
corrupt-hivetrace1,66762.15%68.99%0.43%
redact-ru9,23064.12%69.04%9.36%
synth-ru-tickets32,09967.32%81.51%
synth-wiki-tables13,09468.24%69.54%
synth-jira-comments2,70571.65%72.31%
hivetrace1,66772.17%74.81%0.26%
jayguard1,19573.14%85.77%1.29%
nerel24,36973.30%76.97%
russian-pii-66k4,80575.48%77.79%
factrueval7,96677.54%78.22%
alrosait1,86286.63%99.03%6.07%