Calibration
GPQA Diamond → HLE
HLE is estimated from GPQA Diamond with a inverse Michaelis–Menten curve fitted on 35 models measured on both: y = 0.42396·(x − 0.5946) / (0.5946 + 0.6262 − x), R² = 0.52, cross-validated error 9.2 pp. It is used for 7 estimates.
| Estimated model | GPQA Diamond | HLE | Source |
|---|---|---|---|
| LLaDA2.2-mini | 44.4% | 0.0% | estimated ± 9.2 pp, low confidence |
| Mercury 2.5 | 79.0% | 19.2% | estimated ± 9.2 pp, medium confidence |
| MiniMax M2.7 | 87.0% | 33.3% | estimated ± 9.2 pp, medium confidence |
| Pareto 26.10 Preview | 92.4% | 47.0% | estimated ± 9.2 pp, medium confidence |
| Solar Pro 4 | 89.0% | 37.8% | estimated ± 9.2 pp, medium confidence |
| Trinity-Large-Preview | 63.3% | 2.8% | estimated ± 9.2 pp, low confidence |
| Trinity-Large-Thinking | 76.3% | 15.6% | estimated ± 9.2 pp, medium confidence |