Calibration
MMLU-Redux → HLE
HLE is estimated from MMLU-Redux with a linear curve fitted on 7 models measured on both: y = 2.2056·x + -1.7778, R² = 0.74, cross-validated error 5.9 pp. It is used for 4 estimates.
| Estimated model | MMLU-Redux | HLE | Source |
|---|---|---|---|
| Mellum2-12B-A2.5B-Instruct | 78.1% | 0.0% | estimated ± 5.9 pp, low confidence |
| Mellum2-12B-A2.5B-Thinking | 86.2% | 12.3% | estimated ± 5.9 pp, low confidence |
| MiniCPM5-1B | 70.1% | 0.0% | estimated ± 5.9 pp, low confidence |
| Ternary Bonsai 2 27B | 89.1% | 18.7% | estimated ± 5.9 pp, low confidence |