Calibration
SuperGPQA → MMLU-Redux
MMLU-Redux is estimated from SuperGPQA with a Michaelis–Menten curve fitted on 9 models measured on both: y = 1.1441·x / (0.14541 + x), R² = 0.99, cross-validated error 0.9 pp. It is used for 2 estimates.
| Estimated model | SuperGPQA | MMLU-Redux | Source |
|---|---|---|---|
| Seed 2.1 Pro | 70.8% | 94.9% | estimated ± 0.9 pp, high confidence |
| Seed 2.1 Turbo | 67.4% | 94.1% | estimated ± 0.9 pp, high confidence |
Fitted on
| Model | SuperGPQA | MMLU-Redux |
|---|---|---|
| Claude Opus 4.5 | 70.6% | 96.6% |
| Kimi K2.5 | 69.2% | 94.5% |
| MiniCPM5-1B | 23.1% | 70.1% |
| MiniCPM5-2B | 40.8% | 84.7% |
| Qwen3.5 397B | 70.4% | 94.9% |
| Qwen3.6-27B | 66.0% | 93.5% |
| Qwen3.6 Plus | 71.6% | 94.5% |
| Qwen3.7 Max | 73.6% | 95.0% |
| Qwen3.7 Plus | 71.4% | 94.5% |