EXE -Bench

A comprehensive benchmark ranking AI-based Windows malware detectors across four real-world deployment axes. Drag the sliders to reweight the metrics and instantly see how the leaderboard shifts.

P  Performance — F1 at deployment
T  Temporal — stability over time
R  Robustness — vs adversarial EXEmples
I  Inference — speed on endpoint CPU
Code GitHub

Metric weights

Weights are normalized to sum to 1. Score = w P ·P + w T ·T + w R ·R + w I ·I

Performance (P)
25%
F1 score on test set at deployment time (1% FPR threshold)
Temporal (T)
25%
Sample-weighted F1 across quarterly bins 2019–2022
Robustness (R)
25%
AUC of detection rate vs. adversarial byte-injection budget
Inference (I)
25%
e −t where t is avg CPU inference time (seconds)
Category: Feature Feature-based (EMBER GBDT) E2E End-to-end deep networks DRS De-randomized Smoothing RS Randomized Smoothing

Leaderboard

30 models
# Model Score Metric breakdown P T R I