Calibration
CursorBench 3.1 → AA-SciCode
AA-SciCode is estimated from CursorBench 3.1 with a linear curve fitted on 5 models measured on both: y = 0.3711·x + 0.3413, R² = 0.91, cross-validated error 1.7 pp. It is used for 2 estimates.
| Estimated model | CursorBench 3.1 | AA-SciCode | Source |
|---|---|---|---|
| Claude Sonnet 4.6 | 48.8% | 52.2% | estimated ± 1.7 pp, medium confidence |
| Composer 2.5 | 63.2% | 57.6% | estimated ± 1.7 pp, medium confidence |