Calibration
CyberGym → GDPval-AA
GDPval-AA is estimated from CyberGym with a Hill curve fitted on 17 models measured on both: y = 0.2498 + (0.7469 − 0.2498)·x^6.00 / (0.82410^6.00 + x^6.00), R² = 0.82, cross-validated error 5.1 pp. It is used for 7 estimates.
| Estimated model | CyberGym | GDPval-AA | Source |
|---|---|---|---|
| Claude Mythos 5 | 83.8% | 51.1% | estimated ± 5.1 pp, medium confidence |
| Claude Mythos Preview | 83.1% | 50.5% | estimated ± 5.1 pp, medium confidence |
| Claude Opus 4.6 | 66.6% | 35.8% | estimated ± 5.1 pp, medium confidence |
| Claude Sonnet 4.6 | 65.2% | 34.8% | estimated ± 5.1 pp, medium confidence |
| Gemini 3.5 Flash Cyber | 83.2% | 50.5% | estimated ± 5.1 pp, medium confidence |
| Gemini 3.8 Flash Cyber | 86.2% | 53.2% | estimated ± 5.1 pp, medium confidence |
| Fugu Cyber | 86.9% | 53.8% | estimated ± 5.1 pp, medium confidence |