List your company

ResearchFormerly WizWand

Toxicity Classification

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

96Accuracy

6

Apr 8, 2026

90.4Original Accuracy

6

May 6, 2026

EEUCA 2026 (test)

Llama 8B + 5% synth

62.34Macro F1

5

May 11, 2026

JIGSAW Toxicity Dataset Translated FR

94.9Class 0 Precision

5

Apr 21, 2026

RTP-LX French (test)

84.9Class 0 Precision

5

Apr 21, 2026

Civil reconstructed with controlled shortcut injection (test)

54.9MSTPS

5

Apr 15, 2026

JIGSAW Toxicity Dataset Original EN

96.7Class 0 Precision

4

Apr 21, 2026

Toxicity Classification (test)

Template Attack

88.4ASR

4

Feb 26, 2026

English Civil Comments

85.35AUC-ROC

4

Feb 26, 2026

Multilingual Jigsaw

87.94French Accuracy

4

Feb 26, 2026

DICES 3-class (item-level)

67.79Accuracy (Acc)

3

Apr 10, 2026

VOICED item-level (binary)

80Accuracy

3

Apr 10, 2026

Llama-3.2-3B-Instruct (SRD)

83.97Accuracy

2

Feb 26, 2026

98.4AUC

2

Feb 26, 2026

97.1AUC

2

Feb 26, 2026

Yorùbá (Aggregated Stratified K-Fold)

Hybrid TF-IDF + Logistic Regression with Lexicon- and Token-guided Rewriting

83Accuracy

1

Feb 26, 2026

isiXhosa (Aggregated Stratified K-Fold)

Hybrid TF-IDF + Logistic Regression with Lexicon- and Token-guided Rewriting

63Accuracy

1

Feb 26, 2026