Calibration

FrontierSWE v2 → CursorBench 3.2

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

CursorBench 3.2 is estimated from FrontierSWE v2 with a Hill curve fitted on 7 models measured on both: y = 0.6698 + (1.0864 − 0.6698)·x^5.80 / (0.76266^5.80 + x^5.80), R² = 0.37, cross-validated error 4.1 pp. It is used for 2 estimates.

Estimated model	FrontierSWE v2	CursorBench 3.2	Source
Gemini 4 Argon	55.0%	72.4%	estimated ± 4.1 pp, low confidence
GLM-5.3-Flash	18.1%	67.0%	estimated ± 4.1 pp, low confidence