Calibration
FrontierCode 1.1 Extended → Vals SWE-bench
Vals SWE-bench is estimated from FrontierCode 1.1 Extended with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.2085·x / (0.15744 + x), R² = 0.70, cross-validated error 1.4 pp. It is used for 3 estimates.
| Estimated model | FrontierCode 1.1 Extended | Vals SWE-bench | Source |
|---|---|---|---|
| Claude Opus 5.5 | 63.6% | 96.9% | estimated ± 1.4 pp, medium confidence |
| Claude Sonnet 5.5 | 59.1% | 95.4% | estimated ± 1.4 pp, medium confidence |
| GPT-6 Astra | 64.5% | 97.1% | estimated ± 1.4 pp, low confidence |