Span-level PHI extraction on SHIELD ID category 1.0 (test)
96PrecisionFlash Lite
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Flash Litemode=zero-shot extraction2026.05 | 96 | 88 | |
| GPT-OSS 120Bmode=zero-shot extraction2026.05 | 93 | 82 | |
| GPT-OSS 20Bmode=zero-shot extraction2026.05 | 93 | 87 | |
| Gemini Flashmode=zero-shot extraction2026.05 | 85 | 92 |