PII Arena
GLINER2 detector

gliner2-hivetrace-uni-ru PII detection benchmark

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

21.59%fully hidden annotations
23.95%detected / overlapped
1.16%extra masking
4,906CPU characters / second

Strengths

Highest complete-masking rates by normalized data type.

Logins & usernames54.85%5,490 / 10,009
Customer & employee IDs36.17%1,460 / 4,036
Documents & identifiers35.88%3,121 / 8,699

Watch list

Lowest observed complete-masking rates; inspect before deployment.

Network identifiers0.09%20 / 23,092
Other sensitive attributes2.43%27 / 1,113
Phone numbers & email6.65%956 / 14,377

Language coverage

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

LanguageDatasetsFully hidden ↑Detected ↑Extra masking ↓
Russian1921.59%23.95%1.16%

Configuration, source and version

Family
GLINER2
Revision
5644cdb7c953463fa3a370e76617c03c2dd29f92
Eligible datasets
19 / 41
Character F1
0.281
Untouched diagnostic
76.05% · 103,744 / 136,419
Benchmark version
1.0.3 · 2026-09-09
CPU seconds / 10k chars
2.038
GPU seconds / 10k chars
Not measured

Russian labels; source overlap: nemotron-pii, gretel-multi

Results by dataset

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

DatasetGold annotationsFully hidden ↑Detected ↑Extra masking ↓
rubai-ru3,4583.59%32.07%
alexen21,2615.15%16.26%0.00%
synth-wiki-tables13,0945.74%6.02%
synth-secrets-ru58111.70%15.49%3.05%
redact-ru9,23016.67%19.75%0.80%
nerel24,36917.73%18.64%
jayguard1,19518.08%24.77%0.94%
scanpatch8,70819.09%20.64%0.54%
corrupt-hivetrace1,66721.54%26.09%1.83%
synth-ru-tickets32,09922.02%23.57%
synth-jira-comments2,70524.44%25.25%
corrupt-redmadrobot5,53124.46%27.41%0.53%
alrosait1,86228.30%32.44%1.22%
hivetrace1,66728.97%29.09%0.24%
redmadrobot5,51631.45%32.47%0.27%
russian-pii-66k4,80533.88%35.46%
nym-ru9,49735.60%37.18%
factrueval7,96638.14%38.51%
multiconer-ru1,20839.57%53.06%7.52%