Calibration
Agents' Last Exam → CyberGym
CyberGym is estimated from Agents' Last Exam with a Michaelis–Menten curve fitted on 8 models measured on both: y = 1.6011·x / (0.24083 + x), R² = 0.49, cross-validated error 5.4 pp. It is used for 1 estimate.
| Estimated model | Agents' Last Exam | CyberGym | Source |
|---|---|---|---|
| Qwen3.8-Flash-Next | 51.2% | 100.0% | estimated ± 5.4 pp, low confidence |