List your company

ResearchFormerly WizWand

Watermark Detection

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

Llama2-7B Clean

100F1 Score

8

Apr 3, 2026

OpenGen OPT-6.7B

100TPR

8

Mar 17, 2026

OpenGen Llama-3.1-8B

100TPR

8

Mar 17, 2026

C4 Llama-3.1-8B

100TPR

8

Mar 17, 2026

99.3Detection Rate

7

May 25, 2026

98.1Detection Rate

7

May 25, 2026

87.74Delta Cosine (%)

7

Feb 26, 2026

LLaMA-2 Token Replacement Attack, epsilon=0.2 (1,000 generated sequences)

88.79TPR @ FPR=0.1%

7

Feb 26, 2026

LLaMA-2 Token Replacement Attack epsilon=0.1 (1,000 generated sequences)

96.07TPR@FPR=0.1%

7

Feb 26, 2026

LLaMA-2 Token Replacement Attack epsilon=0.05 (1,000 generated sequences)

97.11TPR@FPR=0.1%

7

Feb 26, 2026

Agent Guide simulation environment Round 2

1.41Z-statistic (Original)

6

Jun 2, 2026

Agent Guide simulation environment Round 1

1.41Z-statistic (Original)

6

Jun 2, 2026

Llama3-8B generated text max 30 tokens

0.008Detection Time (s)

6

Feb 26, 2026

Llama-3-8B Paraphrase perturbation, 150 tokens 1.0 (test)

0.041Mean P

6

Feb 26, 2026

Llama-3-8B Paraphrase perturbation, 30 tokens 1.0 (test)

0.21Mean P

6

Feb 26, 2026

Llama-3-8B Translate perturbation, 150 tokens 1.0 (test)

0.093Mean P

6

Feb 26, 2026

Llama-3-8B Translate perturbation, 30 tokens 1.0 (test)

0.13Mean P

6

Feb 26, 2026

Llama-3-8B Delete perturbation, 150 tokens 1.0 (test)

0.29Mean P

6

Feb 26, 2026

Llama-3-8B Delete perturbation, 30 tokens 1.0 (test)

0.2Mean P

6

Feb 26, 2026

Llama-3-8B Swap perturbation, 150 tokens 1.0 (test)

0.01Mean P

6

Feb 26, 2026

Llama-3-8B Swap perturbation, 30 tokens 1.0 (test)

0.11Mean P

6

Feb 26, 2026

C4 DIPPER attack (test)

1,270TPR @ FPR=0.1%

6

Feb 26, 2026

C4 GPT Rephrase attack (test)

16.4TPR (at FPR=0.1%)

6

Feb 26, 2026

C4 GPT Back Translation attack (test)

58.9TPR (FPR=0.1%)

6

Feb 26, 2026

C4 Random Token Replacement attack (test)

87.5TPR (FPR=0.1%)

6

Feb 26, 2026