List your company

ResearchFormerly WizWand

Input Moderation

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

ToxicChat (test)

Qwen3Guard-4B-Gen-loose

82.8F1 Score

42

May 26, 2026

Qwen3Guard-8B-Gen-strict

91.4F1 Score

26

Apr 14, 2026

WildGuard (test)

Llama3-StreamGuard-8B

89.5F1 Score

22

Apr 7, 2026

HarmBench (test)

Qwen3Guard-4B-Gen-strict

100F1 Score

22

Apr 7, 2026

PolyGuard-Qwen-7B

100F1 Score

22

Apr 7, 2026

AEGIS 2.0 (test)

Llama3-StreamGuard-8B

87.9F1 Score

22

Apr 7, 2026

Input Moderation Benchmark Suite (ToxicChat, OAIMod, Aegis, Aegis2, SSTest, HarmB, WildG)

Llama3-StreamGuard-8B

88.2Macro-average F1

22

Apr 7, 2026

Harmful safety datasets Average

88.33Average F1 Score (Input Moderation)

9

Jun 2, 2026