Calibration
WideResearch → Toolathlon
Toolathlon is estimated from WideResearch with a Hill curve fitted on 7 models measured on both: y = 0.1948 + (1.2000 − 0.1948)·x^6.00 / (0.94486^6.00 + x^6.00), R² = 0.71, cross-validated error 5.2 pp. It is used for 2 estimates.
| Estimated model | WideResearch | Toolathlon | Source |
|---|---|---|---|
| Hy4 preview | 83.9% | 52.5% | estimated ± 5.2 pp, low confidence |
| Qwen3.8 Max | 81.9% | 49.4% | estimated ± 5.2 pp, low confidence |
Fitted on
| Model | WideResearch | Toolathlon |
|---|---|---|
| Claude Opus 4.5 | 76.4% | 43.5% |
| GLM-5 | 69.8% | 38.0% |
| Kimi K2.6 | 80.8% | 50.0% |
| Kimi K2.5 | 72.7% | 27.8% |
| Qwen3.5 397B | 74.0% | 36.3% |
| Qwen3.6-35B-A3B | 60.1% | 26.9% |
| Qwen3.6 Plus | 74.3% | 39.8% |