Calibration

ExploitGym → GDPval-AA

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

GDPval-AA is estimated from ExploitGym with a Michaelis–Menten + offset curve fitted on 14 models measured on both: y = 0.3171 + 0.2458·x / (0.04187 + x), R² = 0.43, cross-validated error 6.5 pp. It is used for 0 estimates.