Calibration
Toolathlon → JobBench
JobBench is estimated from Toolathlon with a Michaelis–Menten curve fitted on 5 models measured on both: y = 2.0000·x / (2.25199 + x), R² = 0.82, cross-validated error 7.3 pp. It is used for 2 estimates.
| Estimated model | Toolathlon | JobBench | Source |
|---|---|---|---|
| Step 3.7 Flash | 49.5% | 36.0% | estimated ± 7.3 pp, low confidence |
| MiniMax M2.7 | 46.3% | 34.1% | estimated ± 7.3 pp, low confidence |
Fitted on
| Model | Toolathlon | JobBench |
|---|---|---|
| Claude Opus 4.5 | 43.5% | 32.3% |
| GPT-5.4 | 54.6% | 38.9% |
| GPT-5.5 | 55.6% | 42.7% |
| Kimi K2.5 | 27.8% | 8.7% |
| Muse Spark 1.1 | 75.6% | 54.7% |