Calibration
SuperGPQA → GPQA Diamond
GPQA Diamond is estimated from SuperGPQA with a Michaelis–Menten + offset curve fitted on 7 models measured on both: y = 0.0000 + 1.8609·x / (0.82112 + x), R² = 0.87, cross-validated error 12.9 pp. It is used for 2 estimates.
| Estimated model | SuperGPQA | GPQA Diamond | Source |
|---|---|---|---|
| Seed 2.1 Pro | 70.8% | 86.2% | estimated ± 12.9 pp, low confidence |
| Seed 2.1 Turbo | 67.4% | 83.9% | estimated ± 12.9 pp, low confidence |
Fitted on
| Model | SuperGPQA | GPQA Diamond |
|---|---|---|
| Claude Opus 4.6 | 95.0% | 89.2% |
| GLM-5 | 66.8% | 86.0% |
| Kimi K2.5 | 69.2% | 87.6% |
| MiniCPM5-1B | 23.1% | 26.3% |
| MiniCPM5-2B | 40.8% | 70.2% |
| Qwen3.7 Max | 73.6% | 92.4% |
| Qwen3.7 Plus | 71.4% | 90.3% |