Calibration
FrontierMath v2 (Tiers 1-3) → FrontierMath v2 (Tier 4)
FrontierMath v2 (Tier 4) is estimated from FrontierMath v2 (Tiers 1-3) with a Hill curve fitted on 44 models measured on both: y = 0.0117 + (0.8577 − 0.0117)·x^3.93 / (0.56533^3.93 + x^3.93), R² = 0.97, cross-validated error 4.5 pp. It is used for 7 estimates.
| Estimated model | FrontierMath v2 (Tiers 1-3) | FrontierMath v2 (Tier 4) | Source |
|---|---|---|---|
| DeepSeek V3 | 1.7% | 1.2% | estimated ± 4.5 pp, medium confidence |
| GPT-4.1 mini | 4.5% | 1.2% | estimated ± 4.5 pp, high confidence |
| GPT-4.1 nano | 1.0% | 1.2% | estimated ± 4.5 pp, medium confidence |
| GPT-4o | 0.3% | 1.2% | estimated ± 4.5 pp, medium confidence |
| Llama 4 Maverick | 0.7% | 1.2% | estimated ± 4.5 pp, medium confidence |
| Llama 4 Scout | 0.0% | 1.2% | estimated ± 4.5 pp, medium confidence |
| o1 | 9.3% | 1.2% | estimated ± 4.5 pp, high confidence |