PII Arena
GLINER detector

gliner-nvidia+ov100 PII detection benchmark

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

56.87%fully hidden annotations
65.52%detected / overlapped
3.75%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Dates & times89.51%290 / 324
Phone numbers & email85.16%3,873 / 4,548
Logins & usernames77.30%5,848 / 7,565

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Bank accounts & cards20.00%13 / 65
Network identifiers29.24%4,893 / 16,736
Other sensitive attributes45.67%406 / 889

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English564.85%78.30%4.05%
Russian556.01%64.14%1.76%

Configuration, source and version

Family
GLINER
Revision
bd23e8ef4425fd04e34c5204ab49ffaa706eae79
Eligible datasets
10 / 41
Character F1
0.630
Untouched diagnostic
34.48% · 28,600 / 82,941
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
Not measured
GPU seconds / 10k chars
0.339

source overlap: nemotron-pii

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
secrets-issues28845.83%64.93%3.38%
synth-ru-tickets32,09949.83%53.52%
nerel24,36955.21%65.40%
leak-museum10155.45%59.41%2.33%
tab-echr3,83056.92%75.38%
jayguard1,19561.76%75.82%1.56%
redact-ru9,23063.47%86.91%9.50%
dialogpii-en3,08773.31%82.80%8.64%
factrueval7,96673.79%74.92%
creddata77678.61%82.22%6.95%