PII Arena
GLINER detector

gliner-nvidia+sent300 PII detection benchmark

Measured on 10 of 41 retained datasets under the same frozen scoring and masking protocol.

75.06%fully hidden annotations
84.07%detected / overlapped
4.73%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Logins & usernames96.70%7,315 / 7,565
Dates & times90.43%293 / 324
Phone numbers & email85.88%3,906 / 4,548

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Bank accounts & cards23.08%15 / 65
Other sensitive attributes50.28%447 / 889
Documents & identifiers59.79%638 / 1,067

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English570.45%83.51%5.11%
Russian575.56%84.13%2.24%

Configuration, source and version

Family
GLINER
Revision
bd23e8ef4425fd04e34c5204ab49ffaa706eae79
Eligible datasets
10 / 41
Character F1
0.756
Untouched diagnostic
15.93% · 13,215 / 82,941
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
Not measured
GPU seconds / 10k chars
0.391

source overlap: nemotron-pii

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
secrets-issues28842.36%60.42%4.45%
leak-museum10156.44%61.39%3.70%
jayguard1,19562.68%77.07%1.98%
tab-echr3,83065.59%83.37%
redact-ru9,23065.89%89.75%12.39%
nerel24,36966.15%77.35%
dialogpii-en3,08777.52%86.72%11.20%
creddata77678.61%82.86%7.85%
factrueval7,96680.86%81.99%
synth-ru-tickets32,09984.64%88.45%