Calibration

LiveCodeBench Pro → SWE-bench Pro

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

SWE-bench Pro is estimated from LiveCodeBench Pro with a offset logistic curve fitted on 6 models measured on both: y = 0.5253 + (0.7666 − 0.5253) / (1 + exp(−99.04·(x − 0.8881))), R² = 1.00, cross-validated error 4.9 pp. It is used for 0 estimates.