Calibration
SuperGPQA → AA-GPQA Diamond
AA-GPQA Diamond is estimated from SuperGPQA with a offset logistic curve fitted on 17 models measured on both: y = 0.7024 + (0.8661 − 0.7024) / (1 + exp(−33.35·(x − 0.5872))), R² = 0.65, cross-validated error 3.7 pp. It is used for 4 estimates.
| Estimated model | SuperGPQA | AA-GPQA Diamond | Source |
|---|---|---|---|
| Seed 2.1 Pro | 70.8% | 86.3% | estimated ± 3.7 pp, high confidence |
| Seed 2.1 Turbo | 67.4% | 85.8% | estimated ± 3.7 pp, high confidence |
| Qwen3 235B 2507 | 62.6% | 83.1% | estimated ± 3.7 pp, high confidence |
| MiniCPM5-1B | 23.1% | 70.2% | estimated ± 3.7 pp, medium confidence |
Fitted on
| Model | SuperGPQA | AA-GPQA Diamond |
|---|---|---|
| Claude Opus 4.5 | 70.6% | 81.0% |
| Claude Opus 4.6 | 95.0% | 84.0% |
| Claude Sonnet 4.6 | 95.0% | 79.9% |
| GLM-5 | 66.8% | 82.0% |
| Kimi K2 | 57.2% | 76.6% |
| Kimi K2.5 | 69.2% | 87.9% |
| MiniCPM5-2B | 40.8% | 70.2% |
| Qwen3.5-122B-A10B | 67.1% | 85.7% |
| Qwen3.5-27B | 65.6% | 85.8% |
| Qwen3.5-35B-A3B | 63.4% | 84.5% |
| Qwen3.5 397B | 70.4% | 89.3% |
| Qwen3.6-27B | 66.0% | 84.2% |
| Qwen3.6-35B-A3B | 64.7% | 84.1% |
| Qwen 3.6 Max (preview) | 73.9% | 88.8% |
| Qwen3.6 Plus | 71.6% | 88.2% |
| Qwen3.7 Max | 73.6% | 92.3% |
| Qwen3.7 Plus | 71.4% | 90.0% |