benchgap
Calibration

ExploitGym → AA AutomationBench

AA AutomationBench is estimated from ExploitGym with a linear curve fitted on 7 models measured on both: y = 0.2810·x + 0.5616, R² = 0.34, cross-validated error 5.5 pp. It is used for 4 estimates.

Estimated modelExploitGymAA AutomationBenchSource
Claude Mythos Preview17.5%61.1%estimated ± 5.5 pp, low confidence
GPT-5.6 Luna12.4%59.6%estimated ± 5.5 pp, low confidence
GPT-5.6 Sol33.7%65.6%estimated ± 5.5 pp, low confidence
GPT-5.6 Terra23.2%62.7%estimated ± 5.5 pp, low confidence