Calibration

ProgramBench → AA Coding Index

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

AA Coding Index is estimated from ProgramBench with a offset logistic curve fitted on 6 models measured on both: y = 0.6809 + (0.7979 − 0.6809) / (1 + exp(−157.29·(x − 0.7727))), R² = 0.62, cross-validated error 10.0 pp. It is used for 0 estimates.