Calibration
ApprenticeBench → CyberGym
CyberGym is estimated from ApprenticeBench with a linear curve fitted on 7 models measured on both: y = 0.7735·x + 0.6707, R² = 0.76, cross-validated error 5.0 pp. It is used for 4 estimates.
| Estimated model | ApprenticeBench | CyberGym | Source |
|---|---|---|---|
| Claude Fable 5 | 34.0% | 93.4% | estimated ± 5.0 pp, low confidence |
| Claude Opus 4.7 | 7.0% | 72.5% | estimated ± 5.0 pp, medium confidence |
| Gemini 3.8 Flash | 24.0% | 85.6% | estimated ± 5.0 pp, medium confidence |
| Grok 4.6 | 13.0% | 77.1% | estimated ± 5.0 pp, medium confidence |