GLINER2 detector
gliner25-fastino+nochunk PII detection benchmark
Measured on 11 of 41 retained datasets under the same frozen scoring and masking protocol.
46.89%fully hidden annotations
49.64%detected / overlapped
3.93%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Phone numbers & email72.82%4,185 / 5,747
Bank accounts & cards72.35%552 / 763
Addresses & locations66.26%10,161 / 15,334
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Logins & usernames11.77%917 / 7,794
Network identifiers12.26%2,145 / 17,498
Customer & employee IDs29.68%702 / 2,365
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 6 | 67.21% | 71.24% | 4.26% |
| Russian | 5 | 42.15% | 44.59% | 1.74% |
Configuration, source and version
- Family
- GLINER2
- Revision
- aaecfe45db1d828c963717054ccb868e8ad1f1d5
- Eligible datasets
- 11 / 41
- Character F1
- 0.562
- Untouched diagnostic
- 50.36% · 46,503 / 92,332
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 0.190
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| secrets-issues | 288 | 19.10% | 34.38% | 2.79% |
| synth-ru-tickets | 32,099 | 19.75% | 20.89% | — |
| factrueval | 7,966 | 43.60% | 44.46% | — |
| leak-museum | 101 | 51.49% | 53.47% | 5.51% |
| tab-echr | 3,830 | 56.19% | 63.05% | — |
| dialogpii-en | 3,087 | 60.87% | 67.74% | 3.19% |
| nerel | 24,369 | 61.01% | 64.61% | — |
| redact-ru | 9,230 | 64.78% | 68.83% | 7.10% |
| nemotron-pii | 9,391 | 74.20% | 75.79% | 0.36% |
| jayguard | 1,195 | 74.64% | 86.78% | 1.61% |
| creddata | 776 | 82.09% | 86.47% | 9.73% |