Calibration
JobBench → CyberGym
CyberGym is estimated from JobBench with a Michaelis–Menten curve fitted on 12 models measured on both: y = 1.5741·x / (0.50519 + x), R² = 0.46, cross-validated error 11.4 pp. It is used for 3 estimates.
| Estimated model | JobBench | CyberGym | Source |
|---|---|---|---|
| Claude 4.1 Opus | 21.9% | 47.6% | estimated ± 11.4 pp, low confidence |
| Claude Haiku 4.5 | 16.0% | 37.9% | estimated ± 11.4 pp, low confidence |
| Qwen3.5 Plus | 18.5% | 42.2% | estimated ± 11.4 pp, low confidence |