Calibration

SciCode → LiveCodeBench v6

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

LiveCodeBench v6 is estimated from SciCode with a offset logistic curve fitted on 11 models measured on both: y = 0.6627 + (0.9055 − 0.6627) / (1 + exp(−42.48·(x − 0.3916))), R² = 0.91, cross-validated error 4.8 pp. It is used for 0 estimates.