GLINER2 detector
gliner25-fastino+sent300 PII detection benchmark
Measured on 11 of 41 retained datasets under the same frozen scoring and masking protocol.
71.99%fully hidden annotations
76.66%detected / overlapped
7.56%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Phone numbers & email95.77%5,504 / 5,747
Logins & usernames94.06%7,331 / 7,794
People's names80.18%19,524 / 24,350
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Dates & times26.29%158 / 601
Customer & employee IDs29.85%706 / 2,365
Other sensitive attributes31.27%278 / 889
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 6 | 71.67% | 77.27% | 8.32% |
| Russian | 5 | 72.06% | 76.51% | 2.56% |
Configuration, source and version
- Family
- GLINER2
- Revision
- aaecfe45db1d828c963717054ccb868e8ad1f1d5
- Eligible datasets
- 11 / 41
- Character F1
- 0.725
- Untouched diagnostic
- 23.34% · 21,554 / 92,332
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 0.133
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| secrets-issues | 288 | 27.43% | 46.18% | 7.31% |
| leak-museum | 101 | 54.46% | 58.42% | 10.53% |
| factrueval | 7,966 | 59.11% | 59.80% | — |
| redact-ru | 9,230 | 66.78% | 73.00% | 6.54% |
| nemotron-pii | 9,391 | 67.77% | 69.93% | 0.36% |
| dialogpii-en | 3,087 | 68.90% | 76.29% | 9.36% |
| synth-ru-tickets | 32,099 | 70.79% | 75.84% | — |
| jayguard | 1,195 | 74.98% | 87.11% | 2.46% |
| nerel | 24,369 | 79.84% | 83.67% | — |
| creddata | 776 | 82.35% | 87.11% | 11.81% |
| tab-echr | 3,830 | 85.09% | 96.92% | — |