Calibration
CursorBench 4.0 → SWE-bench Pro
SWE-bench Pro is estimated from CursorBench 4.0 with a Michaelis–Menten curve fitted on 8 models measured on both: y = 2.0000·x / (0.77968 + x), R² = 0.86, cross-validated error 4.3 pp. It is used for 5 estimates.
| Estimated model | CursorBench 4.0 | SWE-bench Pro | Source |
|---|---|---|---|
| Composer 2.5 | 27.7% | 52.4% | estimated ± 4.3 pp, medium confidence |
| Gemini 3.8 Flash | 39.6% | 67.4% | estimated ± 4.3 pp, high confidence |
| Grok 4.6 | 41.4% | 69.4% | estimated ± 4.3 pp, high confidence |
| Grok 4.7 | 46.3% | 74.5% | estimated ± 4.3 pp, high confidence |
| Muse Spark 1.3 | 41.6% | 69.6% | estimated ± 4.3 pp, high confidence |