Calibration
JobBench → Toolathlon
Toolathlon is estimated from JobBench with a Hill curve fitted on 5 models measured on both: y = 0.2756 + (1.2000 − 0.2756)·x^3.07 / (0.53754^3.07 + x^3.07), R² = 0.99, cross-validated error 7.4 pp. It is used for 3 estimates.
| Estimated model | JobBench | Toolathlon | Source |
|---|---|---|---|
| Claude 4.1 Opus | 21.9% | 33.1% | estimated ± 7.4 pp, low confidence |
| Claude Haiku 4.5 | 16.0% | 29.7% | estimated ± 7.4 pp, low confidence |
| Qwen3.5 Plus | 18.5% | 30.9% | estimated ± 7.4 pp, low confidence |