Calibration
ApprenticeBench → GDP.pdf
GDP.pdf is estimated from ApprenticeBench with a linear curve fitted on 5 models measured on both: y = 0.1012·x + 0.2129, R² = 0.48, cross-validated error 4.4 pp. It is used for 7 estimates.
| Estimated model | ApprenticeBench | GDP.pdf | Source |
|---|---|---|---|
| Claude Opus 4.6 | 5.0% | 21.8% | estimated ± 4.4 pp, low confidence |
| Claude Opus 4.7 | 7.0% | 22.0% | estimated ± 4.4 pp, low confidence |
| Claude Sonnet 4.6 | 2.0% | 21.5% | estimated ± 4.4 pp, low confidence |
| Claude Sonnet 5 | 16.0% | 22.9% | estimated ± 4.4 pp, low confidence |
| GPT-5.4 | 11.0% | 22.4% | estimated ± 4.4 pp, low confidence |
| GPT-5.6 Luna | 7.0% | 22.0% | estimated ± 4.4 pp, low confidence |
| GPT-5.6 Terra | 16.0% | 22.9% | estimated ± 4.4 pp, low confidence |