Calibration

SciCode → SWE-bench Pro

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

SWE-bench Pro is estimated from SciCode with a Hill curve fitted on 15 models measured on both: y = 0.0550 + (0.6977 − 0.0550)·x^6.00 / (0.38871^6.00 + x^6.00), R² = 0.84, cross-validated error 8.1 pp. It is used for 0 estimates.