Calibration

Vals LiveCodeBench → AA Coding Index

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

AA Coding Index is estimated from Vals LiveCodeBench with a offset logistic curve fitted on 39 models measured on both: y = 0.5959 + (0.7724 − 0.5959) / (1 + exp(−172.91·(x − 0.8723))), R² = 0.39, cross-validated error 9.0 pp. It is used for 1 estimate.

Estimated model	Vals LiveCodeBench	AA Coding Index	Source
GLM-4.5	67.4%	59.6%	estimated ± 9.0 pp, low confidence