Calibration
FrontierMath v2 (Tiers 1-3) → AIME26
AIME26 is estimated from FrontierMath v2 (Tiers 1-3) with a Hill curve fitted on 6 models measured on both: y = 0.9537 + (1.2000 − 0.9537)·x^6.00 / (0.67615^6.00 + x^6.00), R² = 0.49, cross-validated error 0.6 pp. It is used for 7 estimates.
| Estimated model | FrontierMath v2 (Tiers 1-3) | AIME26 | Source |
|---|---|---|---|
| DeepSeek V3 | 1.7% | 95.4% | estimated ± 0.6 pp, low confidence |
| GPT-4.1 mini | 4.5% | 95.4% | estimated ± 0.6 pp, low confidence |
| GPT-4.1 nano | 1.0% | 95.4% | estimated ± 0.6 pp, low confidence |
| GPT-4o | 0.3% | 95.4% | estimated ± 0.6 pp, low confidence |
| Llama 4 Maverick | 0.7% | 95.4% | estimated ± 0.6 pp, low confidence |
| Llama 4 Scout | 0.0% | 95.4% | estimated ± 0.6 pp, low confidence |
| o1 | 9.3% | 95.4% | estimated ± 0.6 pp, low confidence |