Calibration
ExploitGym → Hermes Index
Hermes Index is estimated from ExploitGym with a Michaelis–Menten curve fitted on 6 models measured on both: y = 0.8648·x / (0.18140 + x), R² = 0.87, cross-validated error 5.5 pp. It is used for 7 estimates.
| Estimated model | ExploitGym | Hermes Index | Source |
|---|---|---|---|
| GPT-6 Cyber | 48.5% | 62.9% | estimated ± 5.5 pp, low confidence |
| Claude Mythos 5.2 | 38.1% | 58.6% | estimated ± 5.5 pp, low confidence |
| GPT-5.6 Cyber | 32.1% | 55.3% | estimated ± 5.5 pp, low confidence |
| Claude Mythos 5.1 | 23.4% | 48.7% | estimated ± 5.5 pp, low confidence |
| Claude Mythos Preview | 17.5% | 42.5% | estimated ± 5.5 pp, low confidence |
| MiMo-V2.6-Flash | 6.0% | 21.5% | estimated ± 5.5 pp, low confidence |
| Muse Spark 1.1 | 0.8% | 3.7% | estimated ± 5.5 pp, low confidence |
Fitted on
| Model | ExploitGym | Hermes Index |
|---|---|---|
| Claude Opus 5.5 | 33.3% | 63.3% |
| GLM-5.3 | 15.0% | 39.2% |
| GPT-6.1 Sol | 35.1% | 57.1% |
| GPT-6 Astra | 42.4% | 56.3% |
| GPT-6 Luna | 11.6% | 33.9% |
| GPT-6 Sol | 22.1% | 44.1% |