Calibration
SuperGPQA → C-Eval
C-Eval is estimated from SuperGPQA with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.3120·x / (0.29241 + x), R² = 0.89, cross-validated error 0.7 pp. It is used for 15 estimates.
| Estimated model | SuperGPQA | C-Eval | Source |
|---|---|---|---|
| Claude Opus 4.6 | 95.0% | 100.0% | estimated ± 0.7 pp, low confidence |
| Claude Sonnet 4.6 | 95.0% | 100.0% | estimated ± 0.7 pp, low confidence |
| GLM-5 | 66.8% | 91.3% | estimated ± 0.7 pp, medium confidence |
| Kimi K2.5 | 69.2% | 92.2% | estimated ± 0.7 pp, medium confidence |
| MiniCPM5-1B | 23.1% | 58.0% | estimated ± 0.7 pp, low confidence |
| MiniCPM5-2B | 40.8% | 76.4% | estimated ± 0.7 pp, low confidence |
| Qwen3 235B 2507 | 62.6% | 89.4% | estimated ± 0.7 pp, low confidence |
| Qwen3.5-122B-A10B | 67.1% | 91.4% | estimated ± 0.7 pp, medium confidence |
| Qwen3.5-27B | 65.6% | 90.7% | estimated ± 0.7 pp, medium confidence |
| Qwen3.5-35B-A3B | 63.4% | 89.8% | estimated ± 0.7 pp, low confidence |
| Qwen 3.6 Max (preview) | 73.9% | 94.0% | estimated ± 0.7 pp, low confidence |
| Qwen3.7 Max | 73.6% | 93.9% | estimated ± 0.7 pp, low confidence |
| Qwen3.7 Plus | 71.4% | 93.1% | estimated ± 0.7 pp, medium confidence |
| Seed 2.1 Pro | 70.8% | 92.8% | estimated ± 0.7 pp, medium confidence |
| Seed 2.1 Turbo | 67.4% | 91.5% | estimated ± 0.7 pp, medium confidence |