Calibration
τ³-bench results → Toolathlon
Toolathlon is estimated from τ³-bench results with a Hill curve fitted on 6 models measured on both: y = 0.0449 + (1.2000 − 0.0449)·x^6.00 / (0.80460^6.00 + x^6.00), R² = 0.44, cross-validated error 6.0 pp. It is used for 8 estimates.
| Estimated model | τ³-bench results | Toolathlon | Source |
|---|---|---|---|
| Granite 4.2 30B | 62.0% | 24.5% | estimated ± 6.0 pp, low confidence |
| Granite 4.2 3B | 45.8% | 8.3% | estimated ± 6.0 pp, low confidence |
| Granite 4.2 8B | 58.1% | 18.8% | estimated ± 6.0 pp, low confidence |
| LFM2.5-2.6B | 5.7% | 4.5% | estimated ± 6.0 pp, low confidence |
| Mercury 2.5 | 96.0% | 90.3% | estimated ± 6.0 pp, low confidence |
| MiMo-V2.5-Pro | 72.9% | 45.6% | estimated ± 6.0 pp, low confidence |
| Mistral Medium 3.5 128B | 91.4% | 83.3% | estimated ± 6.0 pp, low confidence |
| Pokee-Isaac 28B | 66.2% | 31.8% | estimated ± 6.0 pp, low confidence |