Calibration

Vals SWE-bench → SciCode

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

SciCode is estimated from Vals SWE-bench with a Michaelis–Menten curve fitted on 8 models measured on both: y = 1.2863·x / (1.20621 + x), R² = 0.31, cross-validated error 4.5 pp. It is used for 1 estimate.

Estimated model	Vals SWE-bench	SciCode	Source
Composer 2.5	79.6%	51.1%	estimated ± 4.5 pp, medium confidence