Calibration

MCP Atlas → GDPval-AA

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

GDPval-AA is estimated from MCP Atlas with a Hill curve fitted on 28 models measured on both: y = 0.2086 + (1.2000 − 0.2086)·x^5.77 / (1.01890^5.77 + x^5.77), R² = 0.41, cross-validated error 10.5 pp. It is used for 0 estimates.