GLINER2 detector
gliner25-fastino+ov100 PII detection benchmark
Measured on 11 of 41 retained datasets under the same frozen scoring and masking protocol.
70.46%fully hidden annotations
77.39%detected / overlapped
6.12%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Phone numbers & email96.15%5,526 / 5,747
People's names75.97%18,499 / 24,350
Passwords, keys & tokens75.40%1,444 / 1,915
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Other sensitive attributes33.41%297 / 889
Customer & employee IDs33.87%801 / 2,365
Dates & times34.11%205 / 601
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 6 | 74.44% | 79.10% | 6.73% |
| Russian | 5 | 69.53% | 76.99% | 2.12% |
Configuration, source and version
- Family
- GLINER2
- Revision
- aaecfe45db1d828c963717054ccb868e8ad1f1d5
- Eligible datasets
- 11 / 41
- Character F1
- 0.747
- Untouched diagnostic
- 22.61% · 20,874 / 92,332
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 0.118
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| secrets-issues | 288 | 36.46% | 54.17% | 5.58% |
| leak-museum | 101 | 53.47% | 55.45% | 8.23% |
| factrueval | 7,966 | 55.82% | 56.50% | — |
| redact-ru | 9,230 | 69.19% | 73.92% | 5.81% |
| synth-ru-tickets | 32,099 | 69.97% | 82.61% | — |
| dialogpii-en | 3,087 | 70.10% | 77.26% | 6.65% |
| nemotron-pii | 9,391 | 71.23% | 73.26% | 0.36% |
| nerel | 24,369 | 73.30% | 76.98% | — |
| jayguard | 1,195 | 74.73% | 86.61% | 2.03% |
| creddata | 776 | 82.09% | 86.60% | 10.91% |
| tab-echr | 3,830 | 87.68% | 95.87% | — |