benchgap
Calibration

Vals SWE-bench → CursorBench 3.2

CursorBench 3.2 is estimated from Vals SWE-bench with a Michaelis–Menten curve fitted on 17 models measured on both: y = 2.0000·x / (1.95432 + x), R² = 0.47, cross-validated error 5.4 pp. It is used for 4 estimates.

Estimated modelVals SWE-benchCursorBench 3.2Source
Gemini 3.1 Flash-Lite62.8%48.6%estimated ± 5.4 pp, low confidence
Gemini 3 Flash75.0%55.5%estimated ± 5.4 pp, low confidence
GPT-5.2-Codex72.4%54.1%estimated ± 5.4 pp, low confidence
Qwen3.5 Flash64.4%49.6%estimated ± 5.4 pp, low confidence