Calibration
ExploitBench Internal Port → Chess Puzzles
Chess Puzzles is estimated from ExploitBench Internal Port with a Hill curve fitted on 6 models measured on both: y = 0.3604 + (1.2000 − 0.3604)·x^0.46 / (0.57325^0.46 + x^0.46), R² = 1.00, cross-validated error 5.3 pp. It is used for 3 estimates.
| Estimated model | ExploitBench Internal Port | Chess Puzzles | Source |
|---|---|---|---|
| GPT-6.2 Sol | 50.5% | 76.8% | estimated ± 5.3 pp, low confidence |
| GPT-6 Cyber | 42.0% | 75.0% | estimated ± 5.3 pp, low confidence |
| GPT-6.1 Terra | 24.5% | 69.9% | estimated ± 5.3 pp, low confidence |
Fitted on
| Model | ExploitBench Internal Port | Chess Puzzles |
|---|---|---|
| GPT-5.6 Sol | 5.5% | 59.0% |
| GPT-6.1 Sol | 21.5% | 68.0% |
| GPT-6.2 Astra | 53.5% | 78.0% |
| GPT-6 Astra | 39.0% | 74.0% |
| GPT-6 Luna | 0.0% | 36.0% |
| GPT-6 Sol | 5.5% | 56.0% |