benchgap
Calibration

CursorBench 4.0 → FrontierSWE v2

FrontierSWE v2 is estimated from CursorBench 4.0 with a Hill curve fitted on 8 models measured on both: y = 0.0000 + (0.8214 − 0.0000)·x^6.00 / (0.46480^6.00 + x^6.00), R² = 0.87, cross-validated error 7.6 pp. It is used for 5 estimates.

Estimated modelCursorBench 4.0FrontierSWE v2Source
Claude Sonnet 534.1%11.1%estimated ± 7.6 pp, low confidence
Composer 2.527.7%3.5%estimated ± 7.6 pp, low confidence
GPT-5.6 Luna35.9%14.4%estimated ± 7.6 pp, low confidence
GPT-5.6 Terra41.3%27.1%estimated ± 7.6 pp, medium confidence
Muse Spark 1.341.6%27.9%estimated ± 7.6 pp, medium confidence