List your company

Benign Task Completion on Terminal Bench (held-out)

75.2Trajectory Pass RateBenign Solver
64.38467.1927072.808Jun 8, 2026

Evaluation results

MethodLinks
Benign SolverDefense State=Original, Filtering=LLM Judge-filtered2026.0675.283.1——
Benign SolverDefense State=Hardened, Filtering=LLM Judge-filtered, Method=hacker-fixer loop2026.0664.872.7-10.50.015