Calibration
WideResearch → GDPval-AA
GDPval-AA is estimated from WideResearch with a Michaelis–Menten + offset curve fitted on 5 models measured on both: y = 0.0000 + 2.0000·x / (5.77294 + x), R² = 0.38, cross-validated error 3.7 pp. It is used for 10 estimates.
| Estimated model | WideResearch | GDPval-AA | Source |
|---|---|---|---|
| Atria Dawn Preview | 81.9% | 24.8% | estimated ± 3.7 pp, low confidence |
| Claude Opus 4.5 | 76.4% | 23.4% | estimated ± 3.7 pp, low confidence |
| dots3-note Preview | 78.9% | 24.0% | estimated ± 3.7 pp, low confidence |
| GLM-5 | 69.8% | 21.6% | estimated ± 3.7 pp, low confidence |
| Hy4 preview | 83.9% | 25.4% | estimated ± 3.7 pp, low confidence |
| Ornith-1.5-35B-A3B | 67.8% | 21.0% | estimated ± 3.7 pp, low confidence |
| Ornith-1.5-397B | 80.8% | 24.6% | estimated ± 3.7 pp, low confidence |
| Ornith-1.5-9B | 59.5% | 18.7% | estimated ± 3.7 pp, low confidence |
| Qwen3.5 397B | 74.0% | 22.7% | estimated ± 3.7 pp, low confidence |
| Qwen3.8 Max | 81.9% | 24.8% | estimated ± 3.7 pp, low confidence |