Malicious PTM Detection on Malicious PTM Detection Dataset Benign Malicious 2025 benchmark leaderboard · SOTA2 Research

Malicious PTM Detection on Malicious PTM Detection Dataset Benign Malicious 2025

2,025Benign CountFickling

Evaluation Results

Method	Links
Fickling2026.04		2,025	25	1,000	1,000	2,025	0	2,025	0	50	100	66.67
Llama-3.12026.04		1,337	22	993	995	2,010	688	1,337	15	60.05	99.26	74.83
GPT-5.22026.04		12	20	988	796	1,804	2,013	12	221	99.33	89.09	93.93
DynaHug (default)2026.04		1	25	1,000	986	2,011	2,024	1	14	99.95	99.31	99.63
PickleScan2026.04		0	23	1,000	44	1,067	2,025	0	958	100	52.69	69.02
ModelScan2026.04		0	23	1,000	44	1,067	2,025	0	958	100	52.69	69.02
ModelTracer2026.04		0	25	828	907	1,760	2,025	0	265	100	86.91	92.99