List your company

Verifier Robustness Evaluation on KernelBench task 001 (held-out)

62Hinted ASRNo defense
-2.4814.263147.74Jun 8, 2026

Evaluation results

MethodLinks
No defenseAdversary Model=Gemini 3.1 Pro2026.06627698
Blind, no poolAdversary Model=Gemini 3.1 Pro, Robust at=Iter. 22026.06522098
No defenseAdversary Model=Claude Opus 4.72026.06446190
Blind, no poolAdversary Model=Claude Opus 4.7, Robust at=Iter. 22026.06446190
Verifier-aware + pool (with autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 5, Autopatch=true2026.06416198
Verifier-aware + pool (with autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 2, Autopatch=true2026.0617298
Verifier-aware + pool (with autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 5, Autopatch=true2026.0615292
Verifier-aware + pool (no autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 5, Autopatch=false2026.061520
Verifier-aware + pool (with autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 2, Autopatch=true2026.0611290
Blind + poolAdversary Model=Gemini 3.1 Pro, Robust at=Iter. 52026.066098
Verifier-aware + pool (no autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 5, Autopatch=false2026.06500
Blind + poolAdversary Model=Claude Opus 4.7, Robust at=Iter. 52026.062690
Aware, no poolAdversary Model=Gemini 3.1 Pro, Robust at=Iter. 82026.062098
Verifier-aware + pool (with autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 8, Autopatch=true2026.060098
Verifier-aware + pool (with autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 8, Autopatch=true2026.060092
Verifier-aware + pool (with autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 11, Autopatch=true2026.060098
Verifier-aware + pool (with autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 11, Autopatch=true2026.060092
Verifier-aware + pool (with autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 13 (final), Autopatch=true2026.060098
Verifier-aware + pool (with autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 13 (final), Autopatch=true2026.060092
Verifier-aware + pool (no autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 8, Autopatch=false2026.06000
Verifier-aware + pool (no autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 8, Autopatch=false2026.06000
Verifier-aware + pool (no autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 11, Autopatch=false2026.060094
Verifier-aware + pool (no autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 11, Autopatch=false2026.060022
Verifier-aware + pool (no autopatch)Adversary Model=Gemini 3.1 Pro, Iteration=Iter. 13 (final), Autopatch=false2026.06000
Verifier-aware + pool (no autopatch)Adversary Model=Claude Opus 4.7, Iteration=Iter. 13 (final), Autopatch=false2026.06000
Aware, no poolAdversary Model=Claude Opus 4.7, Robust at=Iter. 82026.060090