Calibration
GPQA → MMLU
MMLU is estimated from GPQA with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (0.9076 − 0.0000)·x^6.00 / (0.35840^6.00 + x^6.00), R² = 0.80, cross-validated error 3.6 pp. It is used for 22 estimates.
| Estimated model | GPQA | MMLU | Source |
|---|---|---|---|
| Claude 3.5 Sonnet | 59.4% | 86.6% | estimated ± 3.6 pp, medium confidence |
| Claude Mythos 5 | 94.1% | 90.5% | estimated ± 3.6 pp, low confidence |
| Claude Sonnet 4.5 | 83.4% | 90.2% | estimated ± 3.6 pp, low confidence |
| Hy4 preview | 92.3% | 90.4% | estimated ± 3.6 pp, low confidence |
| Interfaze Beta | 89.9% | 90.4% | estimated ± 3.6 pp, low confidence |
| LFM2.5-230M | 25.4% | 10.2% | estimated ± 3.6 pp, low confidence |
| LFM2.5-VL-450M | 25.7% | 10.8% | estimated ± 3.6 pp, low confidence |
| MAI-Thinking-1 | 84.2% | 90.2% | estimated ± 3.6 pp, low confidence |
| Mellum2-12B-A2.5B-Instruct | 40.9% | 62.5% | estimated ± 3.6 pp, low confidence |
| Mellum2-12B-A2.5B-Thinking | 57.6% | 85.8% | estimated ± 3.6 pp, medium confidence |
| Ornith-1.5-35B-A3B | 89.2% | 90.4% | estimated ± 3.6 pp, low confidence |
| Ornith-1.5-397B | 92.8% | 90.5% | estimated ± 3.6 pp, low confidence |
| Ornith-1.5-9B | 86.4% | 90.3% | estimated ± 3.6 pp, low confidence |
| Qwen3 235B 2507 | 77.5% | 89.9% | estimated ± 3.6 pp, low confidence |
| Qwen3.8 Max | 92.6% | 90.5% | estimated ± 3.6 pp, low confidence |
| Qwen3.8-Omni-Flash | 91.0% | 90.4% | estimated ± 3.6 pp, low confidence |
| Sakana Fugu | 95.5% | 90.5% | estimated ± 3.6 pp, low confidence |
| Sakana Fugu-Ultra | 95.5% | 90.5% | estimated ± 3.6 pp, low confidence |
| Soofi S 30B-A3B | 43.4% | 68.9% | estimated ± 3.6 pp, low confidence |
| Ternary Bonsai 2 27B | 85.8% | 90.3% | estimated ± 3.6 pp, low confidence |
| ZAYA1-74B-Preview | 57.3% | 85.6% | estimated ± 3.6 pp, medium confidence |
| ZAYA1-8B | 71.0% | 89.3% | estimated ± 3.6 pp, medium confidence |