Calibration

Vals LiveCodeBench → AA-SciCode

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

AA-SciCode is estimated from Vals LiveCodeBench with a offset logistic curve fitted on 38 models measured on both: y = 0.4961 + (0.6499 − 0.4961) / (1 + exp(−86.21·(x − 0.8883))), R² = 0.38, cross-validated error 4.6 pp. It is used for 1 estimate.

Estimated model	Vals LiveCodeBench	AA-SciCode	Source
GLM-4.5	67.4%	49.6%	estimated ± 4.6 pp, low confidence