Explainability classification on AEGIS 2.0 (test) benchmark leaderboard · SOTA2 Research

Explainability classification on AEGIS 2.0 (test)

79.6Unsafe F1LEG large

Evaluation Results

Method	Links
LEG large2026.01		79.6	—
LEG base2026.01		76.95	—
LEG large2026.01		76.93	—
LEG base2026.01		74.28	—
LEG xs2026.01		72.73	—
Word Baseline large2026.01		70.9	—
Word Baseline large2026.01		69.53	—
LEG xs2026.01		69.49	—
Word Baseline base2026.01		66.91	—
Word Baseline base2026.01		64.06	—
LEG large2026.01		52.77	—
Word Baseline large2026.01		52.03	—
SHAP Baseline large2026.01		45.94	—
LEG base2026.01		45.91	—
Word Baseline base2026.01		45.72	—
SHAP Baseline large2026.01		43.99	—
SHAP Baseline base2026.01		41.07	—
SHAP Baseline large2026.01		40.88	—
SHAP Baseline base2026.01		37.21	—
SHAP Baseline base2026.01		33.12	—
LIME Baseline large2026.01		32.74	—
LIME Baseline base2026.01		32.53	—
LIME Baseline large2026.01		31.74	—
LIME Baseline base2026.01		31.18	—
LIME Baseline base2026.01		29.59	—
LIME Baseline large2026.01		29.34	—
LEG xs2026.01		26.48	—
GPT-4o-mini2026.01		—	53.28
LEG base2026.01		—	60.33
LEG large2026.01		—	66.11