Calibration
MMLU-ProX → NOVA-63
NOVA-63 is estimated from MMLU-ProX with a Michaelis–Menten curve fitted on 7 models measured on both: y = 2.0000·x / (2.09804 + x), R² = 0.47, cross-validated error 1.3 pp. It is used for 9 estimates.
| Estimated model | MMLU-ProX | NOVA-63 | Source |
|---|---|---|---|
| DeepSeek V3 0324 | 70.5% | 50.3% | estimated ± 1.3 pp, low confidence |
| GPT-4.1 | 72.7% | 51.5% | estimated ± 1.3 pp, low confidence |
| GPT-4o | 61.1% | 45.1% | estimated ± 1.3 pp, low confidence |
| Nemotron 3 Ultra | 83.0% | 56.7% | estimated ± 1.3 pp, low confidence |
| Phi-4 | 49.9% | 38.4% | estimated ± 1.3 pp, low confidence |
| Qwen3 235B 2507 | 79.4% | 54.9% | estimated ± 1.3 pp, low confidence |
| Qwen3.5-122B-A10B | 82.2% | 56.3% | estimated ± 1.3 pp, low confidence |
| Qwen3.5-27B | 82.2% | 56.3% | estimated ± 1.3 pp, low confidence |
| Qwen3.5-35B-A3B | 81.0% | 55.7% | estimated ± 1.3 pp, low confidence |