Span-level PHI extraction on SHIELD AGE category 1.0 (test)
89Precision (SHIELD AGE 1.0)Gemini Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini Flashmode=zero-shot extraction2026.05 | 89 | 95 | |
| GPT-OSS 20Bmode=zero-shot extraction2026.05 | 88 | 82 | |
| GPT-OSS 120Bmode=zero-shot extraction2026.05 | 87 | 83 | |
| Flash Litemode=zero-shot extraction2026.05 | 79 | 87 |