ResearchBenchmarksBlack-box system prompt extraction on 41 Models Closed-weightFollow4.8Avg TurnsJUSTASK1.162.1053.053.995Jan 29, 2026Evaluation ResultsMethodMethodLinksAvg TurnsExtraction Success RateJUSTASKModels=12, Primary Ski...Models=12, Primary Skills=H8+H4, L14+L72026.014.81JUSTASKModels=6, Primary Skil...Models=6, Primary Skills=L142026.011.31