Calibration
GPQA Diamond → GPQA
GPQA is estimated from GPQA Diamond with a linear curve fitted on 48 models measured on both: y = 1.0006·x + -0.0001, R² = 1.00, cross-validated error 0.4 pp. It is used for 14 estimates.
| Estimated model | GPQA Diamond | GPQA | Source |
|---|---|---|---|
| A.X K2 | 85.6% | 85.6% | estimated ± 0.4 pp, high confidence |
| K-EXAONE 2.0 | 82.2% | 82.2% | estimated ± 0.4 pp, high confidence |
| LLaDA2.2-mini | 44.4% | 44.4% | estimated ± 0.4 pp, high confidence |
| LongCat-Flash-Lite-Sparse | 69.5% | 69.5% | estimated ± 0.4 pp, high confidence |
| Mercury 2.5 | 79.0% | 79.0% | estimated ± 0.4 pp, high confidence |
| MiniCPM5-1B | 26.3% | 26.3% | estimated ± 0.4 pp, high confidence |
| MiniCPM5-2B | 70.2% | 70.2% | estimated ± 0.4 pp, high confidence |
| Pareto 26.10 Preview | 92.4% | 92.4% | estimated ± 0.4 pp, high confidence |
| Beam | 90.5% | 90.5% | estimated ± 0.4 pp, high confidence |
| Solar Open 2 | 86.3% | 86.3% | estimated ± 0.4 pp, high confidence |
| Solar Pro 4 | 89.0% | 89.0% | estimated ± 0.4 pp, high confidence |
| Step 5 Preview | 93.5% | 93.5% | estimated ± 0.4 pp, high confidence |
| Trinity-Large-Preview | 63.3% | 63.3% | estimated ± 0.4 pp, high confidence |
| Trinity-Large-Thinking | 76.3% | 76.3% | estimated ± 0.4 pp, high confidence |