List your company

ResearchFormerly WizWand

Input prompt classification

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

Llama-Guard-3-8B

100Accuracy

5

May 14, 2026

Evolved Attacker 70B

EvoSafety (Guard)-3B

100Accuracy

5

May 14, 2026

Evolved Attacker 8B

EvoSafety (Guard)-3B

100Accuracy

5

May 14, 2026

Evolved Attacker 4B

EvoSafety (Guard)-3B

100Accuracy

5

May 14, 2026

Skill-Seen Method 70B

EvoSafety (Guard)-3B

100Accuracy (%)

5

May 14, 2026

Skill-Seen Method 8B

EvoSafety (Guard)-3B

100Accuracy

5

May 14, 2026

Skill-Seen Method 4B

EvoSafety (Guard)-3B

100Accuracy

5

May 14, 2026

Skill-Unseen Method (70B)

EvoSafety+ (Guard)-3B

100Accuracy

5

May 14, 2026

Skill-Unseen Method 8B

EvoSafety+ (Guard)-3B

99.93Accuracy

5

May 14, 2026

Skill-Unseen Method 4B

EvoSafety+ (Guard)-3B

100Accuracy

5

May 14, 2026