Calibration
CyberGym → APEX-Agents-AA
APEX-Agents-AA is estimated from CyberGym with a inverse Michaelis–Menten curve fitted on 7 models measured on both: y = 0.50355·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.65, cross-validated error 6.0 pp. It is used for 6 estimates.
| Estimated model | CyberGym | APEX-Agents-AA | Source |
|---|---|---|---|
| Claude Mythos Preview | 83.1% | 35.8% | estimated ± 6.0 pp, low confidence |
| Gemini 3.5 Flash Cyber | 83.2% | 35.9% | estimated ± 6.0 pp, low confidence |
| Gemini 3.8 Flash Cyber | 86.2% | 38.1% | estimated ± 6.0 pp, low confidence |
| Ling 3.1 Flash | 87.9% | 39.5% | estimated ± 6.0 pp, low confidence |
| Muse Spark | 43.5% | 14.0% | estimated ± 6.0 pp, low confidence |
| Fugu Cyber | 86.9% | 38.7% | estimated ± 6.0 pp, low confidence |