Calibration
CyberGym → MCP Atlas
MCP Atlas is estimated from CyberGym with a Hill curve fitted on 11 models measured on both: y = 0.0000 + (0.8073 − 0.0000)·x^6.00 / (0.45864^6.00 + x^6.00), R² = 0.74, cross-validated error 11.4 pp. It is used for 4 estimates.
| Estimated model | CyberGym | MCP Atlas | Source |
|---|---|---|---|
| Claude Mythos Preview | 83.1% | 78.5% | estimated ± 11.4 pp, low confidence |
| Gemini 3.5 Flash Cyber | 83.2% | 78.5% | estimated ± 11.4 pp, low confidence |
| Gemini 3.8 Flash Cyber | 86.2% | 78.9% | estimated ± 11.4 pp, low confidence |
| Fugu Cyber | 86.9% | 79.0% | estimated ± 11.4 pp, low confidence |