PII Arena
HF detector

gravitee-small+cpu-int8 PII detection benchmark

Measured on 9 of 41 retained datasets under the same frozen scoring and masking protocol.

43.54%fully hidden annotations
51.52%detected / overlapped
3.65%extra masking
CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Dates & times83.45%1,553 / 1,861
Phone numbers & email68.85%4,468 / 6,489
Bank accounts & cards68.83%1,294 / 1,880

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Logins & usernames2.96%260 / 8,791
Customer & employee IDs4.12%75 / 1,819
Other sensitive attributes20.09%45 / 224

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
English458.56%66.12%2.99%
Russian434.42%43.43%5.46%
Multilingual171.09%73.30%

Configuration, source and version

Family
HF
Revision
f8c27a85c51c0168f07b9dcf00265bf0a4097939
Eligible datasets
9 / 41
Character F1
0.506
Untouched diagnostic
48.48% · 34,949 / 72,095
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 ↓
secrets-issues28811.46%24.65%3.23%
synth-ru-tickets32,09927.55%36.80%
nym-ru9,49736.20%45.70%
tab-echr3,83040.16%55.69%
ameau012,24246.74%47.68%1.74%
redmadrobot5,51660.68%67.64%4.61%
hivetrace1,66769.65%78.22%11.07%
nemotron-pii9,39170.33%76.05%4.56%
kiji-multi7,56571.09%73.30%