Calibration
AA Coding Index → PostTrainBench
PostTrainBench is estimated from AA Coding Index with a Hill curve fitted on 12 models measured on both: y = 0.0267 + (1.2000 − 0.0267)·x^6.00 / (0.91418^6.00 + x^6.00), R² = 0.47, cross-validated error 5.9 pp. It is used for 4 estimates.
| Estimated model | AA Coding Index | PostTrainBench | Source |
|---|---|---|---|
| Qwen3.8 Max Preview | 71.8% | 25.0% | estimated ± 5.9 pp, low confidence |
| Quasar 438B | 61.2% | 12.4% | estimated ± 5.9 pp, low confidence |
| Apodex 1.1 | 60.8% | 12.0% | estimated ± 5.9 pp, low confidence |
| Apodex 1.1 Mini | 60.8% | 12.0% | estimated ± 5.9 pp, low confidence |
Fitted on
| Model | AA Coding Index | PostTrainBench |
|---|---|---|
| Claude Opus 4.7 | 73.6% | 24.3% |
| Claude Opus 4.8 | 74.3% | 31.0% |
| Claude Opus 5 | 78.0% | 35.7% |
| Gemini 3.1 Pro | 68.8% | 18.1% |
| GLM-5.2 | 68.8% | 30.4% |
| GLM-5.3 | 74.8% | 34.5% |
| GPT-5.4 | 71.1% | 17.8% |
| GPT-5.5 | 74.9% | 23.9% |
| GPT-5.6 Sol | 77.4% | 31.0% |
| GPT-6 Astra | 76.9% | 41.9% |
| Grok 4.5 | 72.5% | 22.6% |
| Kimi K3 | 76.2% | 31.5% |