Calibration
Vals MMLU-Pro → HLE
HLE is estimated from Vals MMLU-Pro with a Hill curve fitted on 28 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.94686^6.00 + x^6.00), R² = 0.36, cross-validated error 9.8 pp. It is used for 7 estimates.
| Estimated model | Vals MMLU-Pro | HLE | Source |
|---|---|---|---|
| Claude Haiku 4.5 | 78.7% | 29.8% | estimated ± 9.8 pp, low confidence |
| Gemini 3.1 Flash-Lite | 86.2% | 43.5% | estimated ± 9.8 pp, low confidence |
| GLM-4.5 | 81.2% | 34.1% | estimated ± 9.8 pp, low confidence |
| Laguna M.1 | 68.8% | 15.4% | estimated ± 9.8 pp, low confidence |
| Laguna XS.2 | 69.1% | 15.7% | estimated ± 9.8 pp, low confidence |
| MiMo-V2.5 | 82.9% | 37.3% | estimated ± 9.8 pp, low confidence |
| Qwen3.5 Flash | 84.1% | 39.5% | estimated ± 9.8 pp, low confidence |